betaFeedback

WorksEngineering

Shifted Interaction Functions: A Unified Framework for Imperfect Actuation and Sensing in Consensus Networks

Abstract

Practical networked robotic systems rarely implement ideal consensus interactions exactly, since local actuation constraints, sensing biases, and heterogeneous interaction maps may shift the input-output relations used by distributed controllers. Rather than focusing solely on whether consensus occurs, such imperfections can fundamentally alter the structure of the asymptotic collective behavior, typically yielding global or clustered velocity alignment, while exact static agreement arises only in special cases. This paper develops a shifted-interaction framework for analyzing collective alignment under imperfect actuation and sensing. Shifted functions provide a compact representation of local actuator and sensor maps, allowing actuation constraints and sensing biases to be represented by horizontally and vertically shifted monotone interactions. The framework distinguishes nodal shifted actuation from edge-wise shifted sensing. For shifted actuation, a Lyapunov analysis establishes global velocity alignment under compatible actuation ranges, including bounded, non-invertible, and deadzone-type maps. For shifted sensing, a maximum-principle argument and a paired-sensitivity condition establish convergence to velocity clusters over edges with infinite accumulated sensitivity; uniform sensing sensitivity yields exponential global velocity alignment. In the linear case, closed-form expressions characterize asymptotic behavior and identify actuation as a special case of sensing. Simulations illustrate static, globally aligned, and clustered velocity-aligned behaviors.

Introduction

Distributed consensus algorithms are a fundamental tool for cooperative control and coordination in multiagent systems. Classical protocols are usually built on idealized interactions in which agents apply unconstrained actuation, exchange symmetric relative-state information, and use coupling functions that vanish at the origin [1]. In robotic implementations, however, actuators may saturate or work only in one direction, sensors may contain calibration biases, and agents or edges may have different local limits and gains. Such heterogeneous imperfections arise in fixed-wing unmanned aerial vehicles without reverse thrust, where nonnegative shifted saturation has recently been validated in outdoor formation-flight experiments [2], in satellites with on-off thrusters [3], and in robotic networks affected by biased or asymmetric relative measurements. Remark 2 explains how far our first-order coordination model covers these systems.

Several strands of the literature address related issues, but mostly as separate problems. Consensus under input saturation [4] and under saturated interconnection states [5] has been studied in both classical and recent settings. Distributed and networked extensions further consider nonidentical bounds, asymmetric limits, and connectivity-preserving interactions [6, 7, 8]. Shifted actuation or sensing ranges can exclude the origin and induce nonzero collective velocities, unlike the origin-admissible saturation models in these studies. State- or output-constrained consensus and safety-aware formulations also show that bounded interactions can change collective behavior [9, 10, 11]. Sensor bias in consensus is usually handled from a compensation viewpoint, where biased relative measurements are estimated or rejected rather than used to characterize the motion they induce [12]. Nonlinear edge-function networks have been analyzed through monotonicity and passivity tools [13]; their reciprocal couplings yield static, possibly clustered, steady states, whereas our distinct, offset oriented maps need not admit equilibria and may produce drifting velocity clusters. Closest in spirit, mismatched relative measurements in rigid formations generate distorted shapes and residual motions [14], [15] and have been exploited for maneuvering [16], [17]. These studies draw on the gradient structure of distance-based formations [14]–[16] or linear displacement-based consensus [17], the latter fitting linear shifted sensing. Unlike these formation-specific mechanisms, we compare shifts acting after nodal aggregation with those acting before aggregation along oriented edges. Our shifted-actuation result and sensing maximum principle include non-invertible maps, whereas our velocity-clustering result assumes C1C^1 sensing and comparable paired sensitivities, which deadzones or saturations need not satisfy. Together, these results motivate a consensus-level description of saturation, bias, unilateral actuation, and shifted sensing within a common interaction model.

This paper develops such a description by modeling actuation and relative sensing through linear or nonlinear horizontally and vertically shifted monotone interaction functions. A conventional consensus coupling is an origin-vanishing map of a state or relative-state error. We generalize this map by allowing horizontal and vertical shifts, so that actuator constraints, one-sided actuation, sensor biases, and asymmetric edge couplings can be represented by the same mathematical object. The framework is applied to two complementary channels. In nodal shifted actuation, the aggregate consensus input is transformed before driving each agent. In edge-wise shifted sensing, each measured relative state is transformed before aggregation. Although the notation starts from an underlying undirected interaction graph, each physical edge is treated through two oriented channels, allowing heterogeneous and effectively directed local processing. Actuator limits lead to range-compatibility checks, sensing nonlinearities lead to responsiveness conditions on active edges, and linear shifted maps provide closed-form predictions of drift and limiting offsets. In the linear case, nodal actuation is representable by oriented-edge sensing, but arbitrary heterogeneous sensing gains need not admit a nodal actuation representation. For nonlinear interactions, the location of a shift may change velocity alignment, the steady-state offset, and whether the limiting behavior is static or drifting.

Motivated by these channel-dependent effects, the contributions of this paper are summarized as follows:

  1. A shifted-interaction formulation places actuator saturation, unilateral actuation, sensor bias, dead zones, and asymmetric couplings within a unified consensus model that separates nodal shifted actuation from edge-wise shifted sensing.

  1. For shifted actuation, a range-compatibility condition guarantees global velocity alignment for invertible, bounded, and non-invertible maps, and characterizes the asymptotic motion; static consensus is a nongeneric special case.

  1. For shifted sensing, a maximum-principle analysis is organized through an active-graph viewpoint. Under a paired-sensitivity condition, individual velocities converge and edges with infinite accumulated sensitivity yield cluster-level alignment; uniform sensing sensitivity yields exponential global alignment.

  1. For linear shifts, closed-form expressions give the drift velocity and the limiting configuration, and linear shifted actuation is shown to be a special case of linear shifted sensing, a correspondence lost for nonlinear interactions.

Section II introduces notation, Section III reviews the consensus problem, and Section IV formulates shifted actuation and sensing. Sections V–VII analyze shifted actuation, shifted sensing, and the linear case, Section VIII presents simulations, and Section IX concludes the paper.

Notation

The nn-dimensional identity matrix is denoted by InI_n, and 1n\mathbf{1}_n represents the all-ones vector. The operator diag⁡(⋅)\operatorname{diag}(\cdot) creates a diagonal matrix with its arguments on the diagonal. For a matrix AA, the notation A†A^{\dagger} stands for its Moore–Penrose pseudoinverse, and ker⁡(A)\ker(A) and im⁡(A)\operatorname{im}(A) denote its kernel and image. The operator col⁡(⋅)\operatorname{col}(\cdot) stacks its arguments into a column vector, and ⊗\otimes denotes the Kronecker product. For a scalar function ff and a value yy, f−1(y):={s∈R:f(s)=y}f^{-1}(y):=\{s\in\mathbb{R}:f(s)=y\} denotes the preimage set, which is a singleton when ff is invertible. We write φ(t)=o(χ(t))\varphi(t)=o(\chi(t)) if φ(t)/χ(t)→0\varphi(t)/\chi(t)\to0 as t→∞t\to\infty.

In this paper, we consider NN agents with MM bidirectional interactions. These interactions are encoded by a graph G=(V,E)\mathcal{G}=(\mathcal{V},\mathcal{E}), where each agent is associated with a node in the set V={1,…,N}\mathcal{V}=\{1,\ldots,N\}, and each bidirectional interaction is represented by an edge in the set E⊆{{i,j}:i,j∈V, i≠j}\mathcal{E}\subseteq\{\{i,j\}:i,j\in\mathcal{V},\ i\ne j\}. The edge set E\mathcal{E} has cardinality MM and consists of unordered pairs {i,j}\{i,j\} when agents ii and jj interact. We index the edges by ℓ\ell and use ℓ∈E\ell\in\mathcal{E} and {i,j}∈E\{i,j\}\in\mathcal{E} interchangeably; agents ii and jj are then neighbors. Although information is exchanged in both directions, each agent may process it differently due to heterogeneous sensing, actuation, or local interaction maps. It is therefore convenient to consider both orientations (i,j)(i,j) and (j,i)(j,i) of each edge separately1. We arbitrarily assign one orientation as the positive one, denoted by ℓi\ell^i, and the other as the negative one, denoted by ℓo\ell^o; the sets Eo\mathcal{E}^o and Eo\mathcal{E}^o collect the positively and negatively oriented edges, and we sometimes omit the superscript of ℓ\ell. For each node ii, Ni\mathcal{N}_i is the set of its neighbors, Ei\mathcal{E}_i is the set of oriented edges with initial node ii, i.e., the interactions from which agent ii receives information, and Eii=Ei∩Ei\mathcal{E}_i^i=\mathcal{E}_i\cap\mathcal{E}^i, Eio=Ei∩Eo\mathcal{E}_i^o=\mathcal{E}_i\cap\mathcal{E}^o; see Figure 1.

Figure 1. Example of a bidirectional graph to model asymmetric interactions. For each bidirectional interaction in E\mathcal{E}, the positive and negative orientations are chosen arbitrarily.

The incidence matrix B∈RN×MB\in\mathbb{R}^{N\times M} is defined by [B]iℓ=+1 (−1)[B]_{i\ell}=+1\ (-1) if node ii is the initial (terminal) node of edge ℓ∈Ei\ell\in\mathcal{E}^i, and 00 otherwise; the Laplacian is L=BBTL=BB^{\mathsf{T}} and the edge Laplacian is Le=BTBL_e=B^{\mathsf{T}}B. We also define the out- and in-incidence matrices Bo,Bi∈RN×MB^o,B^i\in\mathbb{R}^{N\times M}: [Bo]iℓ=+1[B^o]_{i\ell}=+1 if ii is the initial node of edge ℓ∈E\ell\in\mathcal{E} and [Bi]iℓ=−1[B^i]_{i\ell}=-1 if ii is the terminal node, and 00 otherwise for both matrices. Note B=Bo+BiB=B^o+B^i. The superscripts of Bo,BiB^o,B^i refer to the role (out/in) of a node in a positively oriented edge, whereas those of Ei,Eo\mathcal{E}^i,\mathcal{E}^o label the orientation; thus the measurement along ℓi\ell^i, taken by the initial node, enters through BoB^o, and that along ℓo\ell^o through BiB^i. We will always consider the following assumption.

Assumption 1: G\mathcal{G} is connected and the interactions by two neighboring agents are bidirectional.

Remark 1: We stress that Assumption 1 goes beyond a standard undirected graph with symmetric interactions between neighbors. Indeed, it accounts for two distinct oriented sensing channels per edge, rendering the graph directed in practice.

{η∈im⁡(L⊗Id):F∈ker⁡(L⊗Id)}\{\eta\in\operatorname{im}(L \otimes I_d): \mathbf{F} \in\ker(L \otimes I_d)\} is nonempty and bounded. General coupled sensing maps are not covered.

Consensus protocols and paper’s objective

Consensus protocols focus on the following limits of the agents’ states xi∈R,i∈Vx_i \in\mathbb{R}, i \in\mathcal{V}

lim⁡t→∞(xi(t)−xj(t))∀{i,j}∈E,(1)\lim_{t \to\infty}(x_i(t)-x_j(t)) \quad\forall\{i,j\}\in\mathcal{E}, \tag*{(1)}

particularly the system is said to have achieved consensus when the previous limits are all equal to zero. For a connected, balanced, and undirected graph, this is equivalent to the convergence onto the consensus manifold M:=ker⁡(L)=ker⁡(BT)\mathcal{M} := \ker(L) = \ker(B^\mathsf{T}), spanned by 1N\mathbf{1}_N, as described in [1]. Let us consider single-integrator agents

x˙i(t)=ui(t),i∈V,(2)\dot{x}_i(t)=u_i(t), \quad i\in\mathcal{V}, \tag*{(2)}

where ui∈Ru_i \in\mathbb{R} is the control input. The standard consensus algorithm [1] is

ui(t)=−∑j∈Ni(xi(t)−xj(t)),i∈V,(3)u_i(t)=-\sum_{j\in\mathcal{N}_i}(x_i(t)-x_j(t)), \quad i\in\mathcal{V}, \tag*{(3)}

which achieves (average) consensus if G\mathcal{G} is undirected and connected, i.e., lim⁡t→∞xi(t)=1∣V∣∑j∈Vxj(0)\lim_{t\to\infty}x_i(t)=\frac{1}{|\mathcal{V}|}\sum_{j\in\mathcal{V}}x_j(0); importantly, such a consensus implies staticity, i.e., lim⁡t→∞x˙i(t)=0\lim_{t\to\infty}\dot{x}_i(t)=0, ∀i∈V\forall i\in\mathcal{V}.

In multi-robot systems, where xix_i may be a position or an orientation, configurations other than static consensus are also of interest, such as drifting consensus, where consensus is achieved but a residual velocity remains, lim⁡t→∞x˙i(t)≠0\lim_{t\to\infty}\dot{x}_i(t)\ne0. This motivates studying velocity alignment through the limits

lim⁡t→∞(x˙i(t)−x˙j(t))∀{i,j}∈E,(4)\lim_{t\to\infty}(\dot{x}_i(t)-\dot{x}_j(t)) \quad\forall\{i,j\}\in\mathcal{E}, \tag*{(4)}

where agents’ velocities are aligned if all (4) equal zero. This paper studies (1) and (4) when actuation and sensing use heterogeneous shifts of origin-crossing functions, e.g., biased measurements and positive-only actuation.

Remark 2: Higher-Order Alignment - For higher-order systems than (2), velocity alignment (4) can extend to acceleration or higher-order. Studying whether these alignments are nonzero is critical in robotics, as nonstaticity may damage actuators or drive the system out of functioning configurations.

The single integrator (2) models the coordination layer of the agents. For instance, in the fixed-wing formation flights of [2], the consensus input acts on the amplitude of an oscillatory guidance maneuver; this amplitude is always non-negative and drives a positive progression speed along the path, so that the map from the consensus input to the progression speed is a non-negative shifted saturation. For second-order agents, a related mechanism may appear one derivative order higher; Section VIII illustrates this only for linear shifts. Discontinuous on–off thrust does not satisfy Assumption 2. When such agents have a velocity-tracking loop obeying the stated hypotheses, the results apply to that coordination layer. For states xi∈Rdx_i \in\mathbb{R}^d with axis-wise shifts and a common orientation of the agents’ coordinate frames, coordinate-separable nodal actuation and oriented sensing maps each yield dd scalar problems over the same graph, so the corresponding results apply per coordinate. For coupled gradient-type actuation Fi=∇ϕiF_i=\nabla\phi_i, with ϕi\phi_i convex and FiF_i locally Lipschitz, η=(L⊗Id)x\eta=(L\otimes I_d)x and V=∑iϕi(−ηi)V=\sum_i\phi_i(-\eta_i) give V=−FT(L⊗Id)F≤0V=-\mathbf{F}^\mathsf{T}(L\otimes I_d)\mathbf{F}\le0, F:=col⁡(Fi(−ηi))\mathbf{F}:=\operatorname{col}(F_i(-\eta_i)); the proof of Theorem 1 carries over if

Shifted actuation and sensing modeling

Practical challenges in consensus implementation

Physically implementing a consensus algorithm requires each agent to sense the relative states of its neighbors, process them into a control input, and actuate this input.

Consider, as in Figure 2, neighbors ii and jj connected by the edge ℓ∈E\ell\in\mathcal{E}, with ℓi=(i,j)\ell^i=(i,j) and ℓo=(j,i)\ell^o=(j,i). Agents ii and jj only sense zℓi=xi−xjz_{\ell^i}=x_i-x_j and zℓo=xj−xiz_{\ell^o}=x_j-x_i, respectively. Note that we can say that zℓiz_{\ell^i} (zℓoz_{\ell^o}) is the ℓi\ell^i-th (ℓo\ell^o-th) relative state concerning the set Ei\mathcal{E}^i (Ej\mathcal{E}^j), since ℓo=(j,i)\ell^o=(j,i) has agent jj as its initial node. The sensing produces the measurements z~ℓi=ψℓi(zℓi)\widetilde{z}_{\ell^i}=\psi_{\ell^i}(z_{\ell^i}) and z~ℓo=ψℓo(zℓo)\widetilde{z}_{\ell^o}=\psi_{\ell^o}(z_{\ell^o}), for some functions ψℓi,ψℓo:R→R\psi_{\ell^i},\psi_{\ell^o}:\mathbb{R}\to\mathbb{R} modeling imperfections. Following (3), each agent aggregates the measurements over Ei=Eii∪Eio\mathcal{E}_i=\mathcal{E}_i^i\cup\mathcal{E}_i^o, i.e.,

Figure 2. Schematic representation of the implementation of a consensus algorithms, such as (3), involving the ℓ\ell-th edge {i,j}∈E\{i,j\}\in\mathcal{E}. Dashed lines indicate bidirectional measurements.

ui=−∑ℓ∈Eiiz~ℓi−∑ℓ∈Eioz~ℓo.(5)u_i=-\sum_{\ell\in\mathcal{E}_i^i}\widetilde{z}_{\ell^i}-\sum_{\ell\in\mathcal{E}_i^o}\widetilde{z}_{\ell^o}. \tag*{(5)}

The input (5) is then executed by the actuation module. With perfect measurements, z~ℓ=zℓ\widetilde{z}_\ell=z_\ell, and faithful actuation, x˙i=ui\dot{x}_i=u_i, the standard protocol (3) is recovered. Physical multi-robot systems, however, face misaligned or miscalibrated components; high-precision hardware mitigates these effects but limits scalability. Conversely, understanding these imperfections allows exploiting shifted sensing or actuation intentionally.

Formalization of consensus with shifted functions

We model deterministic, usually unknown, and heterogeneous imperfections as

x˙i=Fi(ui),z~ℓ=ψℓ(zℓ),(6)\dot{x}_i=F_i(u_i),\quad\widetilde{z}_\ell=\psi_\ell(z_\ell), \tag*{(6)}

for some functions Fi,ψℓ:R→RF_i,\psi_\ell:\mathbb{R}\to\mathbb{R} and for all nodes i∈Vi\in\mathcal{V} and oriented edges ℓ∈Ei∪Eo\ell\in\mathcal{E}^i\cup\mathcal{E}^o. These functions are called shift functions and are detailed in Subsection IV-C.

As in [13], we describe the closed loop compactly in Figure 3: first-order systems Σi\Sigma_i at the nodes, with null references ri=0r_i=0, and measurement systems Ψℓ\Psi_\ell on the oriented edges ℓ∈Ei∪Eo\ell\in\mathcal{E}^i\cup\mathcal{E}^o, whose direction coincides with (is opposite to) that of BB for ℓ∈Ei\ell\in\mathcal{E}^i (ℓ∈Eo\ell\in\mathcal{E}^o). Stacking the signals, Let x=[x1,…,xN]⊤x = [x_1,\ldots,x_N]^\top, y=xy = x, zi=B⊤yz^{\mathrm{i}} = B^\top y, zo=−B⊤yz^{\mathrm{o}} = -B^\top y, z~i=[z~i1,…,z~iMi]⊤\tilde z^{\mathrm{i}} = [\tilde z^{1}_{\mathrm{i}},\ldots,\tilde z^{M_{\mathrm{i}}}_{\mathrm{i}}]^\top and z~o=[z~o1,…,z~oMo]⊤\tilde z^{\mathrm{o}} = [\tilde z^{1}_{\mathrm{o}},\ldots,\tilde z^{M_{\mathrm{o}}}_{\mathrm{o}}]^\top, and u=[u1,…,uN]⊤u = [u_1,\ldots,u_N]^\top, the input in Figure 3 is

Figure 3. Control-theoretical description of a distributed consensus protocol in order to analyze and accommodate shift functions in actuation and relative sensing.

u=−Boz~i+Biz~o.(7)u = -B^{\mathrm{o}}\tilde z^{\mathrm{i}} + B^{\mathrm{i}}\tilde z^{\mathrm{o}}. \tag*{(7)}

Finally, the systems in Figure 3 are modelled by

Σi:{x˙i=Fi(ui),yi=xi,Ψℓ: z~ℓ=ψℓ(zℓ),(8)\Sigma_i:\begin{cases} \dot{x}_i = F_i(u_i),\\ y_i = x_i, \end{cases} \qquad \Psi_\ell:\ \tilde z_\ell= \psi_\ell(z_\ell), \tag*{(8)}

for some shift functions Fi,ψℓ:R→RF_i,\psi_\ell:\mathbb{R}\to\mathbb{R}, for all nodes i∈Vi\in\mathcal{V} and oriented edges ℓ∈Ei∪Eo\ell\in\mathcal{E}^{\mathrm{i}}\cup\mathcal{E}^{\mathrm{o}}. Indeed, the standard protocol (3) with (2) is recovered when all shift functions are the identity.

Shift Functions

The term shift function refers to shifting, vertically and horizontally, the non-decreasing origin-vanishing functions used in consensus, as illustrated in Figure 4. For example, a miscalibrated sensor might shift input and output as sft⁡(s)=af(s−h)+b\operatorname{sft}(s) = af(s-h)+b, where a>0a>0 is a scaling factor, h∈Rh\in\mathbb{R} a horizontal shift, b∈Rb\in\mathbb{R} a vertical shift, and f:R→Rf:\mathbb{R}\to\mathbb{R} is an increasing function vanishing at the origin. The lowercase symbol sft always denotes a scalar local shift function, such as sft⁡i\operatorname{sft}_i at a node or sft⁡ij\operatorname{sft}_{ij} on an oriented measurement; in particular, sft⁡i≡Fi\operatorname{sft}_i\equiv F_i and sft⁡ℓ≡ψℓ\operatorname{sft}_\ell\equiv\psi_\ell in (8), and the figures keep the labels FiF_i for the actuation maps. The uppercase symbol SFT denotes the stacked vector map, e.g., SFT⁡(u)=col⁡(sft⁡1(u1),…,sft⁡N(uN))\operatorname{SFT}(u)=\operatorname{col}(\operatorname{sft}_1(u_1),\ldots,\operatorname{sft}_N(u_N)).

Figure 4. The left subfigure shows six shifted sensing functions associated with three bidirectional interactions, one color per interaction, with solid and dashed lines for its two orientations. The nominal sensing model ψℓ\psi_\ell in (8) is a tanh-type sigmoid, but each ψℓ\psi_\ell, for ℓ∈Ei∪Eo\ell\in\mathcal{E}^{\mathrm{i}}\cup\mathcal{E}^{\mathrm{o}}, differs by scaling and by horizontal and vertical shifts representing sensing imperfections. The right subfigure shows three shifted actuation models FiF_i in (8). These actuation functions are piecewise linear and non-negative by design, with flat intervals representing jammed or saturated actuation.

Assumption 2: Monotone Shift Functions - All shift functions sft⁡(⋅):R→R\operatorname{sft}(\cdot):\mathbb{R}\to\mathbb{R} are locally Lipschitz continuous and monotonically non-decreasing.

Assumption 2 covers linear, sigmoid-like, and piecewise monotone maps with dead zones or flat saturations, and the value at zero need not be zero. Local Lipschitz regularity supplies classical well-posedness and the almost-everywhere differentiability needed in subsequent arguments; the comparison proof of Theorem 2(i) itself uses only continuity and monotonicity. The role of shift parameters aa, hh, and bb is made explicit in Subsection V-D and Section VII.

Consensus under Shifted Actuation

Requirements for the analysis of shifted actuation

Let Ri:=sft⁡i(R)R_i:=\operatorname{sft}_i(\mathbb{R}), an interval by Assumption 2.

Assumption 3: Compatible Actuation Ranges - The ranges of all shift functions are compatible, i.e., the set R:=⋂i∈VRi\mathcal{R}:=\bigcap_{i\in\mathcal{V}}R_i is nonempty and has non-zero length.

Assumption 3 is a sufficient range-compatibility condition for velocity alignment with finite compatible inputs; asymptotic alignment may also occur at range-closure endpoints. The non-zero length condition is discussed after Theorem 1.

Shifted Actuation System Description

For shifted actuation, (8) becomes

Σi:{x˙i=sft⁡i(ui),yi=xi,Ψℓ: z~ℓ=zℓ,(9)\Sigma_i:\begin{cases} \dot{x}_i = \operatorname{sft}_i(u_i),\\ y_i = x_i, \end{cases} \qquad \Psi_\ell:\ \tilde z_\ell= z_\ell, \tag*{(9)}

which with (7) gives u=−(Bo+Bi)B⊤y=−Lxu=-(B^{\mathrm{o}}+B^{\mathrm{i}})B^\top y=-Lx and

x˙=SFT⁡(−Lx),(10)\dot{x}=\operatorname{SFT}(-Lx), \tag*{(10)}

We define η:=Lx∈im⁡(L)=span⁡(1N)⊥\eta:=Lx\in\operatorname{im}(L)=\operatorname{span}(\mathbf{1}_N)^\perp, a subspace isomorphic to RN−1\mathbb{R}^{N-1}.

Shifted Actuation Analysis

We recall LaSalle’s Invariance Principle [18] as a lemma and three auxiliary lemmas.

Lemma 1: LaSalle Invariance Principle [18], Theorem 4.4) - Let Ω⊆D\Omega\subseteq\mathcal{D} be a compact set that is positively invariant with respect to an autonomous first-order system x˙=f(x)\dot{x}=f(x) with f:D→Rnf:\mathcal{D}\to\mathbb{R}^n locally Lipschitz. Let V:D→RV:\mathcal{D}\to\mathbb{R} be a continuously differentiable function such that V˙(x)≤0\dot V(x)\leq0 in Ω\Omega. Let EE be the set of all points in Ω\Omega where V˙(x)=0\dot V(x)=0. Let I\mathcal{I} be the largest invariant set in EE. Then every solution starting in Ω\Omega approaches I\mathcal{I} as t→∞t\to\infty.

Lemma 2: Unique Compatible Velocity - Under Assumptions 2 and 3 there exists exactly one y∗∈Ry^*\in\mathcal{R} such that ∑i∈Vηi=0\sum_{i\in\mathcal{V}}\eta_i=0, ηi∈sft⁡i−1(y∗)\eta_i\in\operatorname{sft}_i^{-1}(y^*) for some η1,…,ηN∈R\eta_1,\ldots,\eta_N\in\mathbb{R}, where sft⁡i−1(y∗)\operatorname{sft}_i^{-1}(y^*) is the preimage set.

Lemma 3: Separable Convexity Criterion. Let f:Rk→Rf:\mathbb{R}^{k}\to\mathbb{R} be a continuously differentiable function for some kk. Suppose that each component of the gradient ∂f∂xi\frac{\partial f}{\partial x_i} is a non-decreasing function of xix_i alone, then ff is convex.

Lemma 4: Radial Unboundedness Test. Let f:Rk→Rf:\mathbb{R}^{k}\to\mathbb{R} be a continuously differentiable, convex function for some kk. Suppose that its set of critical points X⋆:={ξ∈Rk:∇f(ξ)=0k}\mathcal{X}^{\star}:=\{\xi\in\mathbb{R}^{k}:\nabla f(\xi)=0_k\} is nonempty and bounded. Then each ξ⋆∈X⋆\xi^{\star}\in\mathcal{X}^{\star} is a global minimizer of ff, and ff is radially unbounded.

For the respective proofs, see Appendices A, B and C.

Theorem 1: Shifted Actuation Alignment. Consider the multi-agent system (9) whose graph G\mathcal{G} satisfies Assumption 1, and the sft functions in the actuation satisfy Assumptions 2 and 3. The system achieves velocity alignment, i.e., lim⁡t→∞x˙(t)∈span⁡{1N}\lim_{t\to\infty}\dot{x}(t)\in\operatorname{span}\{1_N\}. More precisely, lim⁡t→∞x˙(t)=y⋆1N\lim_{t\to\infty}\dot{x}(t)=y^{\star}1_N, where y⋆y^{\star} is the unique value given by Lemma 2. Moreover, η(t)=Lx(t)→ηˉ∈im⁡(L)\eta(t)=Lx(t)\to\bar{\eta}\in\operatorname{im}(L) with SFT⁡(−ηˉ)=y⋆1N\operatorname{SFT}(-\bar{\eta})=y^{\star}1_N; this limit may depend on x(0)x(0).

Proof: We show that η=Lx\eta=Lx converges to one equilibrium ηˉ\bar{\eta} satisfying SFT⁡(−ηˉ)=y⋆1N\operatorname{SFT}(-\bar{\eta})=y^{\star}1_N, implying velocity alignment. Consider the evolution of η\eta, which is governed by the autonomous first-order system

η˙=Lx˙=LSFT⁡(−η),(11)\dot{\eta}=L\dot{x}=L\operatorname{SFT}(-\eta), \tag*{(11)}

By construction, η\eta clearly stays in im⁡(L)\operatorname{im}(L). The right-hand side of (11) is locally Lipschitz by Assumption 2, so its solutions exist and are unique.

(Lyapunov function and monotonicity) Consider the following candidate Lyapunov function V:im⁡(L)→RV:\operatorname{im}(L)\to\mathbb{R}

V(η)=∑i∈V∫0−ηisft⁡i(s) ds,(12)V(\eta)=\sum_{i\in\mathcal{V}}\int_{0}^{-\eta_i}\operatorname{sft}_i(s)\,\mathrm{d}s, \tag*{(12)}

which is continuously differentiable because each integrand is continuous. Note that VV is non-increasing along trajectories

V˙(η(t))=−∑i∈Vsft⁡i(−ηi(t))η˙i(t)=−SFT⁡(−η(t))TLSFT⁡(−η(t))≤0,(13)\dot{V}(\eta(t))=-\sum_{i\in\mathcal{V}}\operatorname{sft}_i(-\eta_i(t))\dot{\eta}_i(t) =-\operatorname{SFT}(-\eta(t))^{\mathsf T}L\operatorname{SFT}(-\eta(t))\leq0, \tag*{(13)}

for all t≥0t\geq0, because LL is positive semi-definite.

(Boundedness and non-emptiness of the set EE) We now study the set E={η∈im⁡(L)∣V˙(η)=0}E=\{\eta\in\operatorname{im}(L)\mid\dot{V}(\eta)=0\}. The condition η∈im⁡(L)\eta\in\operatorname{im}(L) implies 1NTη=01_N^{\mathsf T}\eta=0, or, equivalently ∑i∈Vηi=0\sum_{i\in\mathcal{V}}\eta_i=0. On the other hand V˙=0\dot{V}=0 can only be satisfied if SFT⁡(−η)∈ker⁡(L)=span⁡(1N)\operatorname{SFT}(-\eta)\in\operatorname{ker}(L)=\operatorname{span}(1_N), i.e., sft⁡i(−ηi)=y⋆\operatorname{sft}_i(-\eta_i)=y^{\star}, ∀i∈V\forall i\in\mathcal{V} for some y⋆∈Ry^{\star}\in\mathbb{R}. Lemma 2 guarantees that there exist exactly one y⋆∈Ry^{\star}\in\mathbb{R} such that the previous two conditions hold simultaneously. The corresponding values of η\eta at which SFT⁡(−η)=y⋆1N\operatorname{SFT}(-\eta)=y^{\star}1_N may be non-unique, because the shift functions are not necessarily invertible; however, they must remain bounded. Indeed, suppose EE was unbounded, then there would exist a sequence of points in EE with at least two unbounded components, one unbounded above (say, index i0i_0) and the other unbounded below (say, index i1i_1), because ∑i∈Vηi=0\sum_{i\in\mathcal{V}}\eta_i=0. However, due to Assumption 2 (increasingness of the shift functions), this would imply sft⁡i0(−ηi0)≥y⋆\operatorname{sft}_{i_0}(-\eta_{i_0})\geq y^{\star}, for all ηi0∈R\eta_{i_0}\in\mathbb{R}, and sft⁡i1(−ηi1)≤y⋆\operatorname{sft}_{i_1}(-\eta_{i_1})\leq y^{\star}, for all ηi1∈R\eta_{i_1}\in\mathbb{R}, which contradicts Assumption 3, as it would force R⊂Ri0∩Ri1={y⋆}\mathbb{R}\subset R_{i_0}\cap R_{i_1}=\{y^{\star}\} to be empty or have zero length. Hence, EE must be bounded.

(Compactness of the set EE) Furthermore, we show that VV is radially unbounded. First, note that VV is the restriction of the function V~:RN→R\widetilde{V}:\mathbb{R}^{N}\to\mathbb{R} to im⁡(L)\operatorname{im}(L), where V~\widetilde{V} is calculated as VV in (12) but defined on a larger domain. Note that ∂V~∂wi(w)=−sft⁡i(−wi)\frac{\partial\widetilde{V}}{\partial w_i}(w)=-\operatorname{sft}_i(-w_i), i.e., ∇V~(w)=−SFT⁡(−w)\nabla\widetilde{V}(w)=-\operatorname{SFT}(-w) for all w∈RNw\in\mathbb{R}^{N}. Each component of ∇V~\nabla\widetilde{V} is a non-decreasing function of wiw_i alone, so by Lemma 3, V~\widetilde{V} is convex, and so is its restriction VV. Additionally, the gradient of VV within im⁡(L)\operatorname{im}(L) is ∇V(η)=−ΠSFT⁡(−η)\nabla V(\eta)=-\Pi\operatorname{SFT}(-\eta), where Π:=IN−1N1N1NT\Pi:=I_N-\frac{1}{N}1_N1_N^{\mathsf T} is the orthogonal projection of RN\mathbb{R}^{N} onto im⁡(L)\operatorname{im}(L). Note that ∇V(η)=0\nabla V(\eta)=0 if and only if SFT⁡(−η)∈ker⁡(Π)=span⁡(1N)\operatorname{SFT}(-\eta)\in\operatorname{ker}(\Pi)=\operatorname{span}(1_N); therefore, points where the gradient vanishes correspond to solutions to ∑i∈Vηi=0,ηi∈−sft⁡i−1(y⋆)\sum_{i\in\mathcal{V}}\eta_i=0,\eta_i\in-\operatorname{sft}_i^{-1}(y^{\star}), i.e., points in EE. According to Lemma 2 the set EE is non-empty and by a previous argument bounded, thus by Lemma 4, applied on im⁡(L)≅RN−1\operatorname{im}(L)\cong\mathbb{R}^{N-1}, VV is radially unbounded. Finally, we can define the compact and positively invariant set Ω={η∈im⁡(L)∣V(η)≤V(η(0))}\Omega=\{\eta\in\operatorname{im}(L)\mid V(\eta)\leq V(\eta(0))\}; indeed, it is positively invariant by (13) and compact because of the radial unboundedness of VV.

(Largest invariant set) Every e∈Ee\in E is an equilibrium of (11). LaSalle’s principle gives dist⁡(η(t),E)→0\operatorname{dist}(\eta(t),E)\to0. To show pointwise convergence, for e∈Ee\in E set De:=(η−e)TL†(η−e)D_e:=(\eta-e)^{\mathsf T}L^{\dagger}(\eta-e). Since η−e∈im⁡(L)\eta-e\in\operatorname{im}(L) and L†L=ΠL^{\dagger}L=\Pi,

D˙e=2∑i(ηi−ei)(sft⁡i(−ηi)−sft⁡i(−ei))≤0\dot{D}_e=2\sum_i(\eta_i-e_i)\bigl(\operatorname{sft}_i(-\eta_i)-\operatorname{sft}_i(-e_i)\bigr)\leq0

by monotonicity. Choose an accumulation point ηˉ∈E\bar{\eta}\in E, which exists by compactness of Ω\Omega and LaSalle’s principle. Then Dˉηˉ(t)\bar{D}_{\bar{\eta}}(t) is non-increasing and tends to zero along a subsequence, so Dˉηˉ(t)→0\bar{D}_{\bar{\eta}}(t)\to0. Positive definiteness of L†L^{\dagger} on im⁡(L)\operatorname{im}(L) yields η(t)→ηˉ\eta(t)\to\bar{\eta}. The limit may depend on the initial condition when EE is not a singleton; nevertheless, x˙(t)=SFT⁡(−η(t))→y⋆1N\dot{x}(t)=\operatorname{SFT}(-\eta(t))\to y^{\star}1_N. Thus velocity alignment holds. □\square

The non-zero length condition in Assumption 3 prevents upper and lower saturation levels of different agents from coinciding, in which case VV would not be radially unbounded. It also excludes constant actuation, typical of leaders. If exactly one agent i0i_0 has constant actuation x˙i0=va\dot{x}_{i_0}=v_a, with vav_a in the range of every other agent but not one of its saturation levels, the same arguments show that all agents attain the velocity vav_a; the case of a fixed agent, va=0v_a=0, is detailed in Corollary 2.

Global invertibility makes the compatible input vector uniquely determined by the shift maps.

Assumption 4: Globally Invertible Shifts - Shift functions are globally invertible.

Under Theorem 1, let ηˉ:=lim⁡t→∞η(t)∈E\bar{\eta}:=\lim_{t\to\infty}\eta(t)\in E, which may depend on x(0)x(0), if the maps are non-invertible. Then Πx(t)=L†η(t)→L†ηˉ\Pi x(t)=L^{\dagger}\eta(t)\to L^{\dagger}\bar{\eta}. Moreover, without invertibility or a rate assumption,

x(t)=L†ηˉ+y⋆t1N+α∞1N+o(1).x(t)=L^{\dagger}\bar{\eta}+y^{\star}t1_N+\alpha_{\infty}1_N+o(1).

Indeed, set q:=x−L†ηˉ−y⋆t1Nq:=x-L^{\dagger}\bar{\eta}-y^{\star}t1_N. At a maximum of qiq_i, (Lq)i≥0(Lq)_i\geq0 and q˙i=sft⁡i(−ηˉi−(Lq)i)−y⋆≤0\dot{q}_i=\operatorname{sft}_i(-\bar{\eta}_i-(Lq)_i)-y^{\star}\leq0; the reverse inequality holds at a minimum. Thus max⁡iqi\max_i q_i is non-increasing and min⁡iqi\min_i q_i is non-decreasing. As Πq=L†(η−ηˉ)→0\Pi q=L^{\dagger}(\eta-\bar{\eta})\to0, both approach the same constant α∞\alpha_{\infty}. Under Assumption 4, EE is a singleton and ηˉi=−sft⁡i−1(y∗)\bar{\eta}_i=-\operatorname{sft}_i^{-1}(y^*), making the relative offset explicit.

The following corollary characterizes when (static) consensus is achieved.

Corollary 1: Static Consensus Recovery. - In Theorem 1 with the additional Assumption 4, consensus among the states xi(t)x_i(t), ∀i∈V\forall i\in\mathcal{V} is achieved if and only if all shift functions intersect at zero input, i.e., sft⁡i(0)=sft⁡j(0),∀i,j∈V\operatorname{sft}_i(0)=\operatorname{sft}_j(0),\forall i,j\in\mathcal{V}. Moreover, the consensus is static if and only if all shift functions pass through the origin, sft⁡i(0)=0,∀i∈V\operatorname{sft}_i(0)=0,\forall i\in\mathcal{V}.

Proof: (Sufficiency) Suppose sft⁡i(0)=sft⁡j(0)\operatorname{sft}_i(0)=\operatorname{sft}_j(0) for all i,j∈Vi,j\in\mathcal{V}. Then y∗=sft⁡i(0)y^*=\operatorname{sft}_i(0) satisfies the conditions of Lemma 2 with ηi=0\eta_i=0, and by the lemma it is the only such value. Moreover, by global invertibility ηˉ=0N\bar{\eta}=\mathbf{0}_N, implying Πx(t)→L†ηˉ=0N\Pi x(t)\to L^\dagger\bar{\eta}=\mathbf{0}_N, and consensus is achieved eventually. If additionally, all shift functions pass through the origin, i.e., sft⁡i(0)=0,∀i∈V\operatorname{sft}_i(0)=0,\forall i\in\mathcal{V}, then y∗=0y^*=0 and knowing that lim⁡t→∞x˙(t)=y∗1N\lim_{t\to\infty}\dot{x}(t)=y^*\mathbf{1}_N (Theorem 1), then the staticity lim⁡t→∞x˙(t)=0\lim_{t\to\infty}\dot{x}(t)=0 follows.

(Necessity) Suppose consensus is asymptotically achieved, i.e., Πx(t)→0N\Pi x(t)\to\mathbf{0}_N; then η(t)=LΠx(t)→0N\eta(t)=L\Pi x(t)\to\mathbf{0}_N, so ηˉ=0N\bar{\eta}=\mathbf{0}_N. The steady-state velocity of each agent is x˙i=sft⁡i(−ηˉi)=sft⁡i(0),∀i∈V\dot{x}_i=\operatorname{sft}_i(-\bar{\eta}_i)=\operatorname{sft}_i(0),\forall i\in\mathcal{V}. By Theorem 1 velocity alignment is achieved, therefore sft⁡i(0)=sft⁡j(0),∀i,j∈V\operatorname{sft}_i(0)=\operatorname{sft}_j(0),\forall i,j\in\mathcal{V}. If staticity is additionally achieved then it must be x˙i=sft⁡i(0)=0\dot{x}_i=\operatorname{sft}_i(0)=0 for all i∈Vi\in\mathcal{V}. □

Corollary 1 shows that imperfections, even in a single agent, typically yield nonzero velocities and non-consensus states, so static consensus is nongeneric. Conversely, shifted actuation can be designed to induce a predictable collective motion, since known actuation maps determine lim⁡t→∞x˙(t)\lim_{t\to\infty}\dot{x}(t) and lim⁡t→∞Πx(t)\lim_{t\to\infty}\Pi x(t).

An anchored network has a fixed agent i0i_0 with sft⁡i0≡0\operatorname{sft}_{i_0}\equiv0. For i≠i0i\ne i_0, boundedness of sft⁡i−1(0)\operatorname{sft}_i^{-1}(0) excludes an unbounded zero-valued saturation plateau. Under invertibility, set ηˉi=−sft⁡i−1(0)\bar{\eta}_i=-\operatorname{sft}_i^{-1}(0) for i≠i0i\ne i_0 and ηˉi0=−∑i≠i0ηˉi\bar{\eta}_{i_0}=-\sum_{i\ne i_0}\bar{\eta}_i, so ηˉ∈im⁡(L)\bar{\eta}\in\operatorname{im}(L).

Corollary 2: Anchored Network. - Consider system (9) under Assumptions 1 and 2, with sft⁡i0≡0\operatorname{sft}_{i_0}\equiv0. Suppose 0∈Ri0\in R_i and sft⁡i−1(0)\operatorname{sft}_i^{-1}(0) is bounded for all i≠i0i\ne i_0. Then x˙(t)→0N\dot{x}(t)\to\mathbf{0}_N and x(t)x(t) converges to a finite limit, possibly dependent on the initial condition. If all remaining shift functions are globally invertible, then

x(t)→x∗:=L†ηˉ+(xi0(0)−(L†ηˉ)i0)1N.x(t)\to x^*:=L^\dagger\bar{\eta}+\left(x_{i_0}(0)-(L^\dagger\bar{\eta})_{i_0}\right)\mathbf{1}_N.

Moreover, x∗=xi0(0)1Nx^*=x_{i_0}(0)\mathbf{1}_N if and only if sft⁡i(0)=0\operatorname{sft}_i(0)=0 for all i≠i0i\ne i_0.

Proof: Here Assumption 3 fails, since R={0}\mathcal{R}=\{0\}, but the proof of Theorem 1 applies with y∗=0y^*=0. The function VV in (12) is unchanged, except that its i0i_0-th term vanishes, so (13) holds and VV is convex. Since the i0i_0-th entry of SFT⁡\operatorname{SFT} is zero, its critical points in im⁡(L)\operatorname{im}(L) form the set E={η∈im⁡(L):sft⁡i(−ηi)=0, i≠i0}E=\{\eta\in\operatorname{im}(L):\operatorname{sft}_i(-\eta_i)=0,\ i\ne i_0\}. Each ηi\eta_i, i≠i0i\ne i_0, lies in the nonempty bounded set −sft⁡i−1(0)-\operatorname{sft}_i^{-1}(0), and ηi0=−∑i≠i0ηi\eta_{i_0}=-\sum_{i\ne i_0}\eta_i, so EE is nonempty and bounded. Lemmas 4 and 1 then give x˙(t)→0N\dot{x}(t)\to\mathbf{0}_N. The monotone-distance argument of Theorem 1 applies to the compact equilibrium set EE and yields η(t)→e∈E\eta(t)\to e\in E. As the anchor is fixed, x(t)→L†e+[xi0(0)−(L†e)i0]1Nx(t)\to L^\dagger e+[x_{i_0}(0)-(L^\dagger e)_{i_0}]\mathbf{1}_N. Under invertibility, E={ηˉ}E=\{\bar{\eta}\} and e=ηˉe=\bar{\eta}, giving the stated x∗x^*. Finally, L†L^\dagger is injective on im⁡(L)\operatorname{im}(L), which meets span⁡(1N)\operatorname{span}(\mathbf{1}_N) only at origin, so x∗∈span⁡(1N)x^*\in\operatorname{span}(\mathbf{1}_N) if and only if ηˉ=0N\bar{\eta}=\mathbf{0}_N, i.e., sft⁡i(0)=0\operatorname{sft}_i(0)=0 for all i≠i0i\ne i_0; then x∗=xi0(0)1Nx^*=x_{i_0}(0)\mathbf{1}_N. □

Corollary 2 shows that fixing a single agent turns drift into a static distortion. The limiting offsets are determined by an element of EE, and become explicitly computable as L†ηˉL^\dagger\bar{\eta} under invertibility.

Role of the shift parameters

For the parametric family sft⁡i(s)=aifi(s−hi)+bi\operatorname{sft}_i(s)=a_i f_i(s-h_i)+b_i, with ai>0a_i>0 and fif_i continuous, strictly increasing, and fi(0)=0f_i(0)=0, the role of each parameter becomes explicit. The range is Ri=bi+aifi(R)R_i=b_i+a_i f_i(\mathbb{R}): vertical shifts translate the range and the scaling stretches it, whereas horizontal shifts leave it unchanged. Hence horizontal shifts never affect the feasibility condition in Assumption 3, while vertical shifts may violate it when fif_i is bounded. The common velocity y∗y^* is the unique root of

∑i∈V[hi+fi−1(y−biai)]=0,(14)\sum_{i\in\mathcal{V}}\left[h_i+f_i^{-1}\left(\frac{y-b_i}{a_i}\right)\right]=0, \tag*{(14)}

and the limiting aggregate inputs are ηˉi=−hi−fi−1((y∗−bi)/ai)\bar{\eta}_i=-h_i-f_i^{-1}\left((y^*-b_i)/a_i\right), which set the state offset L†ηˉL^\dagger\bar{\eta}. For instance, purely horizontal shifts (bi=0b_i=0, ai=aa_i=a, fi=ff_i=f), (14) gives y∗=af(−hˉ)y^*=af(-\bar{h}) with hˉ:=1N∑ihi\bar{h}:=\frac{1}{N}\sum_i h_i, and ηˉi=hˉ−hi\bar{\eta}_i=\bar{h}-h_i: the average horizontal shift sets the drift velocity through ff, while deviations from the average set the non-consensus offsets independently of ff. With purely vertical shifts (hi=0h_i=0, ai=aa_i=a, fi=ff_i=f), y∗y^* solves ∑if−1((y−bi)/a)=0\sum_i f^{-1}((y-b_i)/a)=0, which differs in general from the average 1N∑ibi\frac{1}{N}\sum_i b_i obtained for linear ff. This asymmetry disappears in the linear case of Section VII.

Remark 3: Infeasible Range Case - If the closures of the ranges RiR_i have no common point, velocity alignment is impossible, since a common limiting velocity would belong to the closure of every RiR_i. The Lyapunov decrease V˙≤0\dot{V}\le0 persists, but the coercivity argument used in Theorem 1 no longer applies. The degenerate case in which RiR_i reduces to a point because one agent is fixed is covered by Corollary 2.

Consensus under shifted sensing

Monotone sensing maps first yield a maximum-principle property; a paired-sensitivity condition then yields velocity clusters over edges with infinite accumulated sensitivity, while uniform slope bounds guarantee global, exponential velocity alignment.

Shifted Sensing System Description

For shifted sensing, (8) becomes

Σi:{x˙i=ui,yi=xi,Ψℓ: zℓ=sft⁡ℓ(zℓ).(15)\Sigma_i:\begin{cases} \dot{x}_i=u_i,\\ y_i=x_i, \end{cases} \qquad \Psi_\ell:\ z_\ell=\operatorname{sft}_\ell(z_\ell). \tag*{(15)}

Consider an edge ℓ\ell whose positive orientation is ℓi=(i,j)\ell^i=(i,j). Agent ii measures zℓi=xi−xj=(BTx)ℓz_{\ell i}=x_i-x_j=(B^\mathsf{T}x)_\ell and agent jj measures zℓo=xj−xi=−(BTx)ℓz_{\ell o}=x_j-x_i=-(B^\mathsf{T}x)_\ell. We write sft⁡ij:=sft⁡i\operatorname{sft}_{ij}:=\operatorname{sft}_i and sft⁡ji:=sft⁡j\operatorname{sft}_{ji}:=\operatorname{sft}_j for the maps applied by agents ii and jj, respectively. Hence, in general sft⁡ij(xi−xj)≠−sft⁡ji(xj−xi)\operatorname{sft}_{ij}(x_i-x_j)\ne-\operatorname{sft}_{ji}(x_j-x_i). By (7) and (15),

x˙=−BoSFT⁡i(BTx)+BiSFT⁡o(−BTx),(16)\dot{x}=-B^{\mathsf{o}}\operatorname{SFT}^{\mathsf{i}}(B^\mathsf{T}x)+B^{\mathsf{i}}\operatorname{SFT}^{\mathsf{o}}(-B^\mathsf{T}x), \tag*{(16)}

where SFTi(z)=[sft1i(z1),⋯ ,sftMii(zMi)]T\mathrm{SFT}^{i}(z)=[\mathrm{sft}^{i}_{1}(z_{1}),\cdots,\mathrm{sft}^{i}_{M_i}(z_{M_i})]^{\mathsf{T}} collects the measurements along Ei\mathcal{E}^{i}, and SFTo\mathrm{SFT}^{o} is defined analogously for Eo\mathcal{E}^{o}. Componentwise,

x˙i=−∑j∈Nisftij(xi−xj),(17)\dot{x}_{i}=-\sum_{j\in\mathcal{N}_{i}}\mathrm{sft}_{ij}(x_i-x_j), \tag*{(17)}

and we denote by g:RN→RNg:\mathbb{R}^{N}\to\mathbb{R}^{N} the right-hand side, x˙=g(x)\dot{x}=g(x).

Shifted Sensing Analysis

The core idea of the upcoming analysis relies on the maximum (minimum) principle, which is a tool in the study of partial differential equations, both in continuous settings [19] and in their discrete formulations [20]. A similar approach has been used in opinion dynamics [21, 22]. Its use in this paper allows us to show how the maximum and minimum velocities in the system approach each other during the evolution, and to concretely characterize this behavior through rigorous bounds.

While Lemma 2 establishes existence and uniqueness of steady-state velocity under shifted actuation, analogous reasoning for shifted sensing is more complex. In shifted actuation, compatibility can be checked through the scalar intersection of nodal input ranges. In shifted sensing, the variables are edge states zℓz_{\ell} constrained by the cycle-compatible subspace z∈im⁡(BT)z\in\operatorname{im}(B^{\mathsf{T}}). Figure 5 illustrates this difference: compatible nodal ranges alone do not guarantee velocity alignment, because all agents must realize the same velocity at one compatible edge-state configuration.

Figure 5. Illustration of why shifted sensing requires stronger conditions than range compatibility alone. Three agents over a triangle graph. Reading row-wise, the four subfigures show the agent states, the state derivatives, the six scalar shifted sensing functions, colored by the agent applying them, and the ranges of the nodal inputs ui=−∑jsftij(⋅)u_i=-\sum_j\mathrm{sft}_{ij}(\cdot). Although these nodal ranges are compatible (gray band), the sensing functions follow sftℓ(zℓ)=tanh⁡(zℓ)+bℓ\mathrm{sft}_{\ell}(z_{\ell})=\tanh(z_{\ell})+b_{\ell} and do not satisfy the uniform sensitivity condition in Assumption 5; hence global velocity alignment is not guaranteed, and in this case the velocities form two clusters, {1,3}\{1,3\} and {2}\{2\}. The numerical trajectories exhibit a bounded x1−x3x_1-x_3 and growing inter-cluster separations, consistent with the asymptotic sensitivity characterization in Theorem 2.

The next result first establishes the maximum principle for monotone sensing. For differentiable maps, let aij(t):=sftij′(xi(t)−xj(t))a_{ij}(t):=\mathrm{sft}'_{ij}(x_i(t)-x_j(t)) denote the oriented sensitivity along an edge. We write Ga\mathcal{G}_a for the graph whose physical edges satisfy ∫0∞aij(t) dt=∞\int_{0}^{\infty}a_{ij}(t)\,\mathrm{d}t=\infty; this makes this definition independent of edge orientation. This graph identifies the interactions with infinite accumulated sensitivity.

Theorem 2: Active-Graph Alignment - Consider the shifted-sensing system (16) over a graph satisfying Assumption 1, with maps satisfying Assumption 2.

(i) Along every solution, the largest velocity v‾(t):=max⁡ix˙i(t)\overline{v}(t):=\max_i\dot{x}_i(t) is non-increasing and the smallest velocity v‾(t):=min⁡ix˙i(t)\underline{v}(t):=\min_i\dot{x}_i(t) is non-decreasing. Thus S(t):=v‾(t)−v‾(t)S(t):=\overline{v}(t)-\underline{v}(t) is non-increasing.

(ii) Suppose in addition that the sensing maps are C1C^1 and satisfy the paired-sensitivity condition, for some K≥1K\geq1,

K−1sftji′(−s)≤sftij′(s)≤Ksftji′(−s),s∈R.(18)K^{-1}\mathrm{sft}'_{ji}(-s)\leq\mathrm{sft}'_{ij}(s)\leq K\mathrm{sft}'_{ji}(-s),\qquad s\in\mathbb{R}. \tag*{(18)}

Then every velocity converges to a finite limit vi∞v_i^{\infty}, with vi∞=vj∞v_i^{\infty}=v_j^{\infty} whenever ii and jj belong to the same connected component of Ga\mathcal{G}_a.

Proof: (i) We use a comparison argument that requires only the continuity and monotonicity in Assumption 2, not local Lipschitz regularity. Let x(t)x(t) and x^(t)\widehat{x}(t) be any two solutions of (16), and set δi(t):=xi(t)−x^i(t)\delta_i(t):=x_i(t)-\widehat{x}_i(t). For every node pp at which δp(t)=max⁡iδi(t)\delta_p(t)=\max_i\delta_i(t), δp−δj≥0\delta_p-\delta_j\geq0 for each neighbor jj. Therefore xp−xj≥x^p−x^jx_p-x_j\geq\widehat{x}_p-\widehat{x}_j, and monotonicity gives

δ˙p=−∑j∈Np[sftpj(xp−xj)−sftpj(x^p−x^j)]≤0.\dot{\delta}_p=-\sum_{j\in\mathcal{N}_p}\left[\mathrm{sft}_{pj}(x_p-x_j)-\mathrm{sft}_{pj}(\widehat{x}_p-\widehat{x}_j)\right]\leq0.

Similarly, δ˙q≥0\dot{\delta}_q\geq0 at every index qq attaining min⁡iδi\min_i\delta_i. As both solutions are continuously differentiable, the upper Dini derivative of max⁡iδi\max_i\delta_i is non-positive, and that of min⁡iδi\min_i\delta_i is non-negative. Consequently, the largest difference between two solutions cannot increase, and the smallest difference cannot decrease.

Now fix ϵ>0\epsilon>0 and compare the solution x(t)x(t) with its time shift x(t+ϵ)x(t+\epsilon), using δi(t)=xi(t+ϵ)−xi(t)\delta_i(t)=x_i(t+\epsilon)-x_i(t). It follows that

max⁡ixi(t+ϵ)−xi(t)ϵ\max_i\frac{x_i(t+\epsilon)-x_i(t)}{\epsilon}

is non-increasing in tt, while the corresponding minimum is non-decreasing. Taking ϵ→0+\epsilon\to0^{+} and using continuity of x˙\dot{x} proves the claim for v‾(t)\overline{v}(t) and v‾(t)\underline{v}(t). This also implies that the velocities remain bounded by their initial extrema; hence trajectories cannot have finite escape time. The comparison applies to arbitrary solutions and also yields uniqueness of solutions with the same initial state.

(ii) Differentiation along solutions gives

v˙i=∑j∈Niaij(t)(vj−vi),aij(t)≥0.\dot{v}_i=\sum_{j\in\mathcal{N}_i}a_{ij}(t)(v_j-v_i),\qquad a_{ij}(t)\geq0.

Condition (18) ensures K−1aij(t)≤aji(t)≤Kaij(t)K^{-1}a_{ij}(t)\leq a_{ji}(t)\leq Ka_{ij}(t) for every physical edge and time. Therefore this induced velocity system is cut-balanced. Applying Theorem 1(a)–(b) of [23] to this continuous-time velocity system yields existence of each vi∞v_i^{\infty} and equality of limiting velocities on every connected component of the graph of infinite accumulated interactions, namely Ga\mathcal{G}_a. □\square

The graph Ga\mathcal{G}_a describes persistent sensitivity, not merely bounded relative distances. In particular, if ∣xi(t)−xj(t)∣\lvert x_i(t)-x_j(t)\rvert remains bounded and sftij′\mathrm{sft}'_{ij} is positive on that bounded interval, then then {i,j}\{i,j\} belongs to Ga\mathcal{G}_a. For sft⁡ij(s)=tanh⁡(s)+bij\operatorname{sft}_{ij}(s)=\tanh(s)+b_{ij}, the paired sensitivities are identical (K=1K=1); thus the two-cluster behavior in Figure 5 is covered without requiring any finite limiting state. Condition (18) is specific to the cluster-convergence claim: it need not hold when the two orientations have different decay rates (e.g., one uses tanh⁡(s)\tanh(s) and the other ss), while non-C1C^1 sensing maps remain covered by part (i). Global alignment is recovered under uniform sensitivity as follows.

Assumption 5: Uniform Sensing Sensitivity. There exist constants 0<κ‾≤κ‾0<\underline{\kappa}\leq\overline{\kappa} such that every oriented sensing function satisfies

κ‾≤sft⁡ij(s)−sft⁡ij(r)s−r≤κ‾for all s≠r.\underline{\kappa}\leq\frac{\operatorname{sft}_{ij}(s)-\operatorname{sft}_{ij}(r)}{s-r}\leq\overline{\kappa}\qquad\text{for all }s\ne r.

For differentiable maps, Assumption 5 reads κ‾≤sft⁡ij′≤κ‾\underline{\kappa}\leq\operatorname{sft}'_{ij}\leq\overline{\kappa}. It is an all-edge-active condition: every sensing map is strictly increasing and onto R\mathbb{R}, e.g., tanh⁡(z−h)+cz+b\tanh(z-h)+cz+b with c>0c>0, which excludes saturated or deadzone sensing, and κ‾\underline{\kappa} prevents the state-dependent coupling weights from vanishing.

Theorem 3: All-Edge Sensing Alignment. Consider the shifted-sensing closed-loop system (16) whose graph G\mathcal{G} satisfies Assumption 1, and whose sensing shift functions satisfy Assumptions 2 and 5. Then the system achieves global velocity alignment exponentially fast: lim⁡t→∞x˙(t)=c1N\lim_{t\to\infty}\dot{x}(t)=c\mathbf{1}_N for some c∈Rc\in\mathbb{R}, and, for any T>0T>0, S(t)≤(1−γ)⌊t/T⌋S(0)S(t)\leq(1-\gamma)^{\lfloor t/T\rfloor}S(0) with γ∈(0,1)\gamma\in(0,1) given in (21), depending only on TT, NN, the maximum graph degree, and the bounds κ‾,κ‾\underline{\kappa},\overline{\kappa}.

Proof: By Assumption 2, the sensing maps are locally Lipschitz; since x(t)x(t) is continuously differentiable, v(t)=g(x(t))v(t)=g(x(t)) is locally Lipschitz on every compact time interval. Thus, for almost every tt we have

ddtsft⁡ij(zij(t))=κij(t)z˙ij(t)\frac{\mathrm{d}}{\mathrm{d}t}\operatorname{sft}_{ij}(z_{ij}(t))=\kappa_{ij}(t)\dot{z}_{ij}(t)

with κij(t)∈[κ‾,κ‾]\kappa_{ij}(t)\in[\underline{\kappa},\overline{\kappa}]. Indeed, if z˙ij(t)≠0\dot{z}_{ij}(t)\ne0, the derivative is the limit of the product of the difference quotient of sft⁡ij\operatorname{sft}_{ij}, which lies in [κ‾,κ‾][\underline{\kappa},\overline{\kappa}] by Assumption 5, and the difference quotient of zijz_{ij}, which tends to z˙ij(t)≠0\dot{z}_{ij}(t)\ne0; if z˙ij(t)=0\dot{z}_{ij}(t)=0, both sides vanish for any κij(t)\kappa_{ij}(t), e.g., κij(t)=κ‾\kappa_{ij}(t)=\underline{\kappa}. Hence, for almost every tt,

v˙i=−∑j∈Niκij(t)(vi−vj),i∈V.(19)\dot{v}_i=-\sum_{j\in\mathcal{N}_i}\kappa_{ij}(t)(v_i-v_j),\qquad i\in\mathcal{V}. \tag*{(19)}

i.e., the velocities obey a consensus protocol over the original bidirectional graph with time-varying arc weights in [κ‾,κ‾][\underline{\kappa},\overline{\kappa}]. To estimate the contraction of the velocity spread, fix t0≥0t_0\geq0 and T>0T>0, let vˉ0:=vˉ(t0)\bar{v}_0:=\bar{v}(t_0), S0:=S(t0)S_0:=S(t_0), and define ei(t):=vˉ0−vi(t)e_i(t):=\bar{v}_0-v_i(t) for t≥t0t\geq t_0. By Theorem 2(i), ei(t)≥0e_i(t)\geq0. From (19), e˙i=∑j∈Niκij(ej−ei)\dot{e}_i=\sum_{j\in\mathcal{N}_i}\kappa_{ij}(e_j-e_i), so that, with a:=dmax⁡κ‾a:=d_{\max}\overline{\kappa} and dmax⁡d_{\max} the maximum degree,

e˙i≥−aei+κ‾ejfor every j∈Ni.(20)\dot{e}_i\geq-ae_i+\underline{\kappa}e_j\qquad\text{for every }j\in\mathcal{N}_i. \tag*{(20)}

Let qq satisfy vq(t0)=vˉ(t0)v_q(t_0)=\bar{v}(t_0), so eq(t0)=S0e_q(t_0)=S_0, and let τ:=T/(N−1)\tau:=T/(N-1). Since eq≥0e_q\geq0 and e˙q≥−κ‾dmax⁡eq\dot{e}_q\geq-\overline{\kappa}d_{\max}e_q, we have eq(t)≥γ0S0e_q(t)\geq\gamma_0S_0 on $[t_0,t_0+T] with γ0:=e−aT\gamma_0:=e^{-aT}. Suppose that an agent jj satisfies ej(s)≥γk−1S0e_j(s)\geq\gamma_{k-1}S_0 for s∈[t0+(k−1)τ,t0+T]s\in[t_0+(k-1)\tau,t_0+T], and let i∈Nji\in\mathcal{N}_j. Multiplying (20) by eate^{at} and integrating from s0:=t0+(k−1)τs_0:=t_0+(k-1)\tau, and using ei≥0e_i\geq0, yields for t∈[t0+kτ,t0+T]t\in[t_0+k\tau,t_0+T]

ei(t)≥κ‾γk−1S0∫s0te−a(t−s) ds≥γkS0,e_i(t)\geq\underline{\kappa}\gamma_{k-1}S_0\int_{s_0}^{t}e^{-a(t-s)}\,\mathrm{d}s\geq\gamma_kS_0,

with γk:=κτe−aTγk−1\gamma_k:=\kappa\tau e^{-aT}\gamma_{k-1}. Since G\mathcal{G} is connected, every agent is at graph distance at most N−1N-1 from qq, so by induction all agents satisfy ei(t0+T)≥γS0e_i(t_0+T)\geq\gamma S_0 with

γ:=min⁡0≤k≤N−1γk,γk=e−aT(κ‾τe−aT)k.(21)\gamma:=\min_{0\leq k\leq N-1}\gamma_k,\qquad\gamma_k=e^{-aT}\left(\underline{\kappa}\tau e^{-aT}\right)^k. \tag*{(21)}

Therefore vˉ(t0+T)=vˉ0−min⁡iei(t0+T)≤vˉ0−γS0\bar{v}(t_0+T)=\bar{v}_0-\min_i e_i(t_0+T)\leq\bar{v}_0-\gamma S_0, and since vˉ(t0+T)≥vˉ(t0)\bar{v}(t_0+T)\geq\bar{v}(t_0) by Theorem 2(i), S(t0+T)≤(1−γ)S0S(t_0+T)\leq(1-\gamma)S_0. Because γ\gamma does not depend on t0t_0, iterating gives S(t)≤(1−γ)⌊t/T⌋S(0)S(t)\leq(1-\gamma)^{\lfloor t/T\rfloor}S(0). Finally, vˉ\bar{v} is non-increasing, is non-decreasing, and their difference vanishes, so both converge to a common limit cc, i.e., x˙(t)→c1N\dot{x}(t)\to c\mathbf{1}_N. □\square

The factor γ\gamma in (21) is conservative: it certifies an exponential rate rather than estimating it.

Remark 4: Sensing Compatibility. - Theorem 2 does not determine the clusters a priori, and their velocities are not given by a scalar intersection of nodal ranges. Writing g(x)=g^(BTx)g(x)=\hat{g}(B^\mathsf{T}x) with g^(z):=−B∘SFT⁡i(z)+BiSFT⁡o(−z)\hat{g}(z):=-B^\circ\operatorname{SFT}^{\mathrm{i}}(z)+B^\mathrm{i}\operatorname{SFT}^{\mathrm{o}}(-z), a coherent constant-velocity motion with fixed finite relative states exists if and only if there are z∈im⁡(BT)z\in\operatorname{im}(B^\mathsf{T}) and c∈Rc\in\mathbb{R} satisfying g^(z)=c1N\hat{g}(z)=c\mathbf{1}_N. This feasibility condition by itself does not establish attraction to that motion. This is the sensing-side analogue of the range-intersection condition for shifted actuation in Assumption 3, but it is more involved because the feasible variable is constrained by the cycle-compatible edge subspace im⁡(BT)\operatorname{im}(B^\mathsf{T}).

Lemma 5: Global LaSalle Criterion ([18], Corollary 4.2). Let x=0x=0 be an equilibrium point of an autonomous first-order system with locally Lipschitz right-hand side. Let V:RN→RV:\mathbb{R}^N\to\mathbb{R} be a continuously differentiable, radially unbounded, positive definite function such that V˙(x)≤0\dot{V}(x)\leq0 for all x∈RNx\in\mathbb{R}^N. Let Z={x∈RN∣V˙(x)=0}\mathcal{Z}=\{x\in\mathbb{R}^N\mid\dot{V}(x)=0\} and suppose that no solution can stay identically in Z\mathcal{Z} other than the trivial solution x(t)≡0x(t)\equiv0. Then, the origin is globally asymptotically stable.

Proposition 1: Odd Sensing Consensus. - Consider the shifted-sensing closed-loop system (16) under Assumptions 1 and 2. Suppose all shift functions are odd and satisfy sft⁡ℓi(s)>0\operatorname{sft}_{\ell^i}(s)>0 for every s≠0s\ne0 and every positive orientation ℓi∈Ei\ell^i\in\mathcal{E}^i. For each edge ℓ∈E\ell\in\mathcal{E}, let its two oriented measurements be ℓi∈Ei\ell^i\in\mathcal{E}^i and ℓo∈Eo\ell^o\in\mathcal{E}^o, and suppose further that sft⁡ℓi(⋅)=sft⁡ℓo(⋅)\operatorname{sft}_{\ell^i}(\cdot)=\operatorname{sft}_{\ell^o}(\cdot). Under these assumptions, the system achieves a static consensus configuration.

Proof: For identical odd sensing maps in both orientations, the global LaSalle criterion in Lemma 5 applies to an edge-potential construction. Let z=BTxz=B^\mathsf{T}x. Under the oddness and edge-wise equality assumptions, (16) reduces to x˙=−BSFT⁡i(z)\dot{x}=-B\operatorname{SFT}^{\mathrm{i}}(z). Hence the edge dynamics are

z˙=−BTBSFT⁡i(z)=−LESFT⁡i(z).(22)\dot{z}=-B^\mathsf{T}B\operatorname{SFT}^{\mathrm{i}}(z)=-L_{\mathcal{E}}\operatorname{SFT}^{\mathrm{i}}(z). \tag*{(22)}

which evolve in im⁡(BT)\operatorname{im}(B^\mathsf{T}). Consider the following Lyapunov-candidate function V:im⁡(BT)→RV:\operatorname{im}(B^\mathsf{T})\to\mathbb{R}

V(z)=∑ℓi∈Ei∫0zℓisft⁡ℓi(s) ds.(23)V(z)=\sum_{\ell^i\in\mathcal{E}^i}\int_0^{z_{\ell^i}}\operatorname{sft}_{\ell^i}(s)\,\mathrm{d}s. \tag*{(23)}

where the sum is performed over the positive edges of the graph. By the strict sign condition, sft⁡ℓi(s)>0\operatorname{sft}_{\ell^i}(s)>0 for s≠0s\ne0, so each term is positive for zℓi≠0z_{\ell^i}\ne0 and, for ∣zℓi∣≥1|z_{\ell^i}|\geq1, satisfies ∫0zℓisft⁡ℓi(s) ds≥cℓi∣zℓi∣\int_0^{z_{\ell^i}}\operatorname{sft}_{\ell^i}(s)\,\mathrm{d}s\geq c_{\ell^i}|z_{\ell^i}|, where ci:=12min⁡{sft⁡ℓi(12),−sft⁡ℓi(−12)}>0c_i := \frac{1}{2}\min\{\operatorname{sft}_{\ell_i}(\frac{1}{2}),-\operatorname{sft}_{\ell_i}(-\frac{1}{2})\}>0, using the positive and negative half-axes separately. Hence VV is positive definite and radially unbounded. Moreover, it satisfies

V˙=∑i∈E∑ℓ∈Eisft⁡i(ziℓ)z˙iℓ=(SFT⁡i(z))⊤z˙.\dot{V}=\sum_{i\in\mathcal{E}}\sum_{\ell\in\mathcal{E}_i}\operatorname{sft}_{i}(z_i^\ell)\dot{z}_i^\ell=(\operatorname{SFT}^{i}(z))^\top\dot{z}.

Therefore,

V˙=−(SFT⁡i(z))⊤B⊤BSFT⁡i(z)=−∥BSFT⁡i(z)∥2=−∥x˙∥2.\dot{V}=-(\operatorname{SFT}^{i}(z))^\top B^\top B\operatorname{SFT}^{i}(z)=-\lVert B\operatorname{SFT}^{i}(z)\rVert^2=-\lVert\dot{x}\rVert^2.

In order to apply LaSalle’s Invariance Principle (Lemma 5), we study the set Z={z∈im⁡(B⊤)∣V˙=0}\mathcal{Z}=\{z\in\operatorname{im}(B^\top)\mid\dot{V}=0\}. Then, for all z∈Zz\in\mathcal{Z} it must be that BSFT⁡i(z)=0NB\operatorname{SFT}^{i}(z)=0_N. We claim this can only happen if z=0Mz=0_M. In fact, since z∈im⁡(B⊤)z\in\operatorname{im}(B^\top), there exists some x∈RNx\in\mathbb{R}^N such that z=B⊤xz=B^\top x; hence, for z∈Zz\in\mathcal{Z}, it must be that z⊤SFT⁡i(z)=x⊤BSFT⁡i(z)=0z^\top\operatorname{SFT}^{i}(z)=x^\top B\operatorname{SFT}^{i}(z)=0. Simultaneously, z⊤SFT⁡i(z)=∑i∈E∑ℓ∈Eiziℓsft⁡iℓ(ziℓ)z^\top\operatorname{SFT}^{i}(z)=\sum_{i\in\mathcal{E}}\sum_{\ell\in\mathcal{E}_i}z_i^\ell\operatorname{sft}_i^\ell(z_i^\ell). By the strict sign condition, if a relative state is nonzero, ziℓ≠0z_i^\ell\ne0, its corresponding term in the sum is strictly positive; hence all relative states are zero, ziℓ=0z_i^\ell=0. Therefore, the set Z\mathcal{Z} contains a single point Z={0M}\mathcal{Z}=\{0_M\}, which is an equilibrium point of the dynamical system. Since im⁡(B⊤)\operatorname{im}(B^\top) is invariant and isomorphic to RN−1\mathbb{R}^{N-1}, the LaSalle criterion applies on this subspace. Application of Lemma 5 shows that the system converges globally asymptotically to z=0Mz=0_M, corresponding to a static consensus equilibrium. □

The strict sign condition is essential for Proposition 1: with the odd deadzone map sft⁡(z)=0\operatorname{sft}(z)=0 for ∣z∣≤1\lvert z\rvert\le1 and sft⁡(z)=z−sign⁡(z)\operatorname{sft}(z)=z-\operatorname{sign}(z) otherwise, every non-consensus configuration with ∣xi−xj∣≤1\lvert x_i-x_j\rvert\le1 on all edges is a static equilibrium. The proof only uses the pairwise antisymmetry sft⁡oℓ(−s)=−sft⁡iℓ(s)\operatorname{sft}_{o}^{\ell}(-s)=-\operatorname{sft}_{i}^{\ell}(s), which may replace oddness and the edge-wise equality. Such homogeneous endpoint measurements are nongeneric; in general, relative-state inconsistencies propagate and drifts occur.

Linear Shift Functions

The linear case admits an explicit solution and approximates the unsaturated regime. With sft⁡(s)=as+b\operatorname{sft}(s)=as+b, a>0a>0, with the appropriate node or oriented-edge sub-index, horizontal and vertical shifts produce identical dynamics, since a(s−h)+b=as+(b−ah)a(s-h)+b=as+(b-ah), unlike the nonlinear case (Subsection V-D). Both (9) and (15) lead to

x˙=−L~x+b,(24)\dot{x}=-\widetilde{L}x+b, \tag*{(24)}

where for the shifted actuation case we have L~=diag⁡(ai)L\widetilde{L}=\operatorname{diag}(a_i)L and b=[b1,…,bN]⊤b=[b_1,\ldots,b_N]^\top with positive slopes ai>0a_i>0, i∈Vi\in\mathcal{V}. In the shifted sensing case, consistent with (16),

L~=BoDiB⊤+BiDoB⊤,b=−Bobi+Bibo,\widetilde{L}=B^{o}D^{i}B^\top+B^{i}D^{o}B^\top,\qquad b=-B^{o}b^{i}+B^{i}b^{o},

where Di=diag⁡(aℓi)D^{i}=\operatorname{diag}(a_{\ell_i}) and Do=diag⁡(aℓo)D^{o}=\operatorname{diag}(a_{\ell_o}) contain the positive slopes aℓi,aℓo>0a_{\ell_i},a_{\ell_o}>0, while bib^{i} and bob^{o} collect the initial-side and terminal-side bias terms, respectively. In either case, L~\widetilde{L} is a weighted Laplacian (zero row sums, non-positive off-diagonal entries) whose kernel is spanned by 1N\mathbf{1}_N, all other eigenvalues having positive real part [24], and its left null vector vv can be chosen positive, so v⊤1N>0v^\top\mathbf{1}_N>0 [1]. Both linear interaction channels admit (24), and Proposition 2 applies to both. Every linear shifted actuation is reproduced by a linear shifted sensing with slopes aℓ=aia_\ell=a_i and biases bℓ=−bi/∣Ei∣b_\ell=-b_i/\lvert\mathcal{E}_i\rvert for ℓ∈Ei\ell\in\mathcal{E}_i. Conversely, since nodal biases can always be distributed in this way, a linear shifted sensing is a linear shifted actuation if and only if each agent

applies a common slope to all its measurements, i.e., aℓ=aia_\ell=a_i for all ℓ∈Ei\ell\in\mathcal{E}_i. Hence linear actuation is a special case of linear sensing; e.g., diag⁡(ai)L\operatorname{diag}(a_i)L has a real spectrum, being similar to diag⁡(ai)1/2Ldiag⁡(ai)1/2\operatorname{diag}(a_i)^{1/2}L\operatorname{diag}(a_i)^{1/2}, while sensing Laplacians may have complex eigenvalues.

Proposition 2: Linear Drift Formula - Consider the multi-agent system (24) under Assumption 1, corresponding to using linear shift functions for FiF_i and ψℓ\psi_\ell in (8). Then

x(t)=P1bt+L~†P2b+c′1N+o(1),t→∞,x(t)=P_1bt+\widetilde{L}^{\dagger}P_2b+c'\mathbf{1}_N+o(1),\qquad t\to\infty,

where P1=11N⊤vv1N⊤P_1=\frac{1}{\mathbf{1}_N^\top v}v\mathbf{1}_N^\top, vv is the left eigenvector of L~\widetilde{L} associated with the zero eigenvalue, P2=I−P1P_2=I-P_1, and c′=v⊤(x(0)−L~†P2b)v⊤1Nc'=\frac{v^\top(x(0)-\widetilde{L}^{\dagger}P_2b)}{v^\top\mathbf{1}_N}. The convergence is exponential.

Proof. Note that P1P_1 and P2P_2 are complementary projections, with L~P1=0\widetilde{L}P_1=0 and v⊤P2=0v^\top P_2=0. Let c:=v⊤bv⊤1Nc:=\frac{v^\top b}{v^\top\mathbf{1}_N}, so that P1b=c1NP_1b=c\mathbf{1}_N. Since v⊤P2b=0v^\top P_2b=0 and im⁡(L~)=ker⁡(v⊤)\operatorname{im}(\widetilde{L})=\ker(v^\top), the vector P2bP_2b lies in im⁡(L~)\operatorname{im}(\widetilde{L}); as L~L~†\widetilde{L}\widetilde{L}^\dagger is the orthogonal projection onto im⁡(L~)\operatorname{im}(\widetilde{L}), the vector x∗:=L~†P2b+c′1Nx^*:=\widetilde{L}^\dagger P_2b+c'\mathbf{1}_N satisfies L~x∗=P2b\widetilde{L}x^*=P_2b. Define ξ(t):=x(t)−c1Nt−x∗\xi(t):=x(t)-c\mathbf{1}_Nt-x^*. Using (24),

ξ˙=−L~(ξ+c1Nt+x∗)+b−c1N=−L~ξ+b−P2b−P1b=−L~ξ.\begin{aligned} \dot{\xi}=-\widetilde{L}(\xi+c\mathbf{1}_Nt+x^*)+b-c\mathbf{1}_N \\ &=-\widetilde{L}\xi+b-P_2b-P_1b=-\widetilde{L}\xi. \end{aligned}

Moreover, v⊤x=v⊤bv^\top x=v^\top b gives v⊤x(t)=v⊤x(0)+(v⊤b)tv^\top x(t)=v^\top x(0)+(v^\top b)t, and the choice of c′c' gives v⊤x∗=v⊤x(0)v^\top x^*=v^\top x(0); hence v⊤ξ(t)=0v^\top\xi(t)=0 for all tt. The subspace ker⁡v⊤=im⁡(L~)\ker v^\top=\operatorname{im}(\widetilde{L}) is invariant under L~\widetilde{L} and complementary to span⁡(1N)\operatorname{span}(\mathbf{1}_N), because v⊤1N≠0v^\top\mathbf{1}_N\ne0, so the restriction of L~\widetilde{L} to it has exactly the nonzero eigenvalues of L~\widetilde{L}, all with positive real part. Therefore ξ(t)→0\xi(t)\to0 exponentially, which proves the claim with lim⁡t→∞x˙(t)=c1N=P1b\lim_{t\to\infty}\dot{x}(t)=c\mathbf{1}_N=P_1b. The constant vector x∗x^* determines the eventual relative states, which are not zero for heterogeneous shift functions in general. □

The bias splits into b∥:=P1bb_{\parallel}:=P_1b, which determines the final velocity, and b⊥:=P2bb_{\perp}:=P_2b, which determines the non-consensus distortion L~†b⊥\widetilde{L}^{\dagger}b_{\perp}. The norm of the distortion is bounded by ∥L~†b⊥∥2≤∥L~†∥2∥b⊥∥2=1σ2(L~)∥b⊥∥2\lVert\widetilde{L}^{\dagger}b_{\perp}\rVert_2\le\lVert\widetilde{L}^{\dagger}\rVert_2\lVert b_{\perp}\rVert_2=\frac{1}{\sigma_2(\widetilde{L})}\lVert b_{\perp}\rVert_2, where σ2(L~)\sigma_2(\widetilde{L}) is the smallest nonzero singular value of L~\widetilde{L}.

Remark 5: Connectivity Distortion Bound - If L~\widetilde{L} is symmetric, then σ2(L~)=λ2(L~)\sigma_2(\widetilde{L})=\lambda_2(\widetilde{L}) is the algebraic connectivity, and the distortion bound reduces to 1/λ21/\lambda_2. This holds, e.g., for identical actuation slopes, or for aℓi=aℓoa_{\ell_i}=a_{\ell_o} on every edge; then better connectivity reduces the maximum distortion.

Simulations

Three scenarios on the graph of Figure 1 illustrate the results: non-negative actuation with flat regions, as in systems without reverse thrust or with jammed actuators; smooth actuator calibration maps, whose drift and limiting offsets are predicted in closed form; and biased nonlinear relative sensing. We close with second-order agents.

Figure 6 illustrates the first example, which uses piecewise linear non-negative actuation. Although these functions are not invertible, their ranges have a common intersection containing the origin, so the range-compatibility condition in Assumption 3 holds and Lemma 2 gives y∗=0y^*=0. Moreover, every FiF_i vanishes exactly on u≤0u\le0 and is positive for u>0u>0, so −Fi−1(0)=[0,∞)-F_i^{-1}(0)=[0,\infty) and the set EE in the proof of Theorem 1 reduces to {η:∑iηi=0, ηi≥0}={0N}\{\eta:\sum_i\eta_i=0,\ \eta_i\geq0\}=\{0_N\}. Hence static consensus is guaranteed by the theory, and the agent initialized at the maximum state remains stationary while the others move toward it.

Figure 6. Piecewise linear non-negative shifted actuation in system (9). From left to right, the three subfigures show the agent states, the state derivatives, and the non-negative shifted actuation functions. The agents reach static consensus at the maximum initial state.

The second example uses the invertible maps of Figure 7, which satisfy Assumption 4. All velocities converge to the same nonzero value, i.e., a drifting rather than static configuration. The dashed reference lines, computed from y∗y^* and L†ηˉL^\dagger\bar{\eta} as in § 4, match the limits of the simulated trajectories.

Figure 7. Smooth invertible shifted actuation with Fi(u)=asinh⁡(u−hi)+biF_i(u)=\operatorname{asinh}(u-h_i)+b_i. Reading row-wise, the four subfigures show the agent states, the state derivatives, the non-consensus components of the state, and the invertible shifted actuation functions. The dashed curves in the third subfigure denote the closed-form predictions for the limiting non-consensus components, and the dashed horizontal line in the second subfigure denotes the aligned velocity.

For comparison, the nonlinear sensing maps in Figure 8 satisfy Assumption 5 with κ=0.2\kappa=0.2 and κˉ=1.2\bar{\kappa}=1.2. Their states do not approach static values, whereas the velocities align, in agreement with Theorem 3.

Figure 8. Nonlinear shifted sensing in system (15) with sft⁡ℓ(z)=tanh⁡(z−hℓ)+0.2z+bℓ\operatorname{sft}_\ell(z)=\tanh(z-h_\ell)+0.2z+b_\ell. The left subfigure shows the agent states and the right subfigure shows the state derivatives. Velocity alignment is achieved under Assumption 5.

To illustrate Remark 2, consider double integrators with linear actuation shifts and the standard second-order protocol, x¨=diag⁡(ai)u+b\ddot{x}=\operatorname{diag}(a_i)u+b with u=−L(x+γx˙)u=-L(x+\gamma\dot{x}), ai>0a_i>0, and γ>0\gamma>0. Since diag⁡(ai)L\operatorname{diag}(a_i)L is similar to diag⁡(ai)1/2Ldiag⁡(ai)1/2\operatorname{diag}(a_i)^{1/2}L\operatorname{diag}(a_i)^{1/2}, each disagreement mode satisfies s2+γλs+λ=0s^2+\gamma\lambda s+\lambda=0 for an eigenvalue λ>0\lambda>0 of diag⁡(ai)L\operatorname{diag}(a_i)L, and is exponentially stable. The left null vector of diag⁡(ai)L\operatorname{diag}(a_i)L is w=diag⁡(ai)−11Nw=\operatorname{diag}(a_i)^{-1}1_N, so w⊤x¨=w⊤bw^\top\ddot{x}=w^\top b, and the argument of Proposition 2 gives x¨(t)→c1N\ddot{x}(t)\to c1_N with c=∑ibi/ai/∑i1/aic=\sum_i b_i/a_i\mathbin{/}\sum_i1/a_i: the shifts produce a common constant acceleration instead of a common velocity.

On the graph of Figure 1 with a=(1,0.8,1.2,1)a=(1,0.8,1.2,1), b=(0.2,−0.1,0.3,0)b=(0.2,-0.1,0.3,0), and γ=1\gamma=1, this predicts c≈0.0796c\approx0.0796, which the accelerations in Figure 9 match to within 10−910^{-9}; consequently, the velocities grow linearly. If instead ∑ibi/ai=0\sum_i b_i/a_i=0, e.g., b=(0.8,−0.96,1.2,−0.6)b=(0.8,-0.96,1.2,-0.6), then w⊤x˙w^\top\dot{x} is constant and agents starting at rest reach a static configuration with offsets L†diag⁡(ai)−1bL^\dagger\operatorname{diag}(a_i)^{-1}b.

Figure 9. Second-order agents with linear actuation shifts. The first two subfigures show the agent states and accelerations for b=(0.2,−0.1,0.3,0)b=(0.2,-0.1,0.3,0), with ∑ibi/ai≠0\sum_i b_i/a_i\ne0: the accelerations approach the predicted value cc (dashed line), so the velocities grow linearly. The last two subfigures show the velocities and the non-consensus components of the state for b=(0.8,−0.96,1.2,−0.6)b=(0.8,-0.96,1.2,-0.6), with ∑ibi/ai=0\sum_i b_i/a_i=0: the agents reach a static configuration with the predicted offsets (dashed lines).

Conclusion

This paper has represented actuator saturation, unilateral actuation, biased measurements, and asymmetric edge interactions as horizontal and vertical shifts of monotone functions, which allows comparing imperfections in the actuation and sensing channels within the same consensus model. For shifted actuation, the analysis gives a range-based check for velocity alignment and includes bounded as well as non-invertible maps. For shifted sensing, the maximum-principle analysis first yields non-increasing velocity spread and cluster convergence under a paired-sensitivity condition, without assuming state convergence, and then recovers global exponential velocity alignment when all sensing edges remain responsive.

The results show that the location and shape of the shift are both essential. Linear shifted interactions lead to the same class of affine closed-loop descriptions for actuation and sensing, actuation being a special case of sensing, whereas nonlinear shifted interactions may generate channel-dependent steady-state offsets, static consensus, or drifting consensus. In practice, actuator ranges indicate whether a common velocity can be realized, sensing responsiveness whether the active graph remains connected, and linear calibration maps anticipate the drift and the limiting configuration. Future work will relax the all-edge-active condition for shifted sensing, especially for saturated and deadzone-type measurement maps, and will extend the proposed framework to higher-order dynamics, coupled multidimensional shifts, and one-way or time-varying interaction graphs.

Proof of Lemma 2

Proof: For existence, we seek a common velocity y∈Ry \in\mathbb{R} and numbers ηi∈−sft⁡i−1(y)\eta_i \in-\operatorname{sft}_i^{-1}(y) satisfying ∑iηi=0\sum_i \eta_i = 0. Equivalently, put ui=−ηiu_i=-\eta_i and H:={u∈RN:1N⊤u=0}H:=\{u\in\mathbb{R}^N: \mathbf{1}_N^\top u=0\}. Consider the continuously differentiable, convex function

W(u)=∑i∈V∫0uisft⁡i(s) ds,u∈H.W(u)=\sum_{i\in\mathcal{V}}\int_0^{u_i}\operatorname{sft}_i(s)\,\mathrm{d}s,\qquad u\in H.

Let a<ba<b lie in the interior of the common range R\mathcal{R} and write c=(a+b)/2c=(a+b)/2. For each ii, choose ri−,ri+∈Rr_i^-,r_i^+\in\mathbb{R} with sft⁡i(ri−)=a\operatorname{sft}_i(r_i^-)=a and sft⁡i(ri+)=b\operatorname{sft}_i(r_i^+)=b. Monotonicity of sft⁡i\operatorname{sft}_i gives, for a suitable finite constant CiC_i,

∫0uisft⁡i(s) ds≥cui+b−a2∣ui∣−Ci,ui∈R.(25)\int_0^{u_i}\operatorname{sft}_i(s)\,\mathrm{d}s\ge cu_i+\frac{b-a}{2}|u_i|-C_i,\qquad u_i\in\mathbb{R}. \tag*{(25)}

Indeed, for ui≥ri+u_i\ge r_i^+ the integrand is at least bb, whereas for ui≤ri−u_i\le r_i^- it is at most aa; on the intervening compact interval the difference between the two sides has a finite lower bound. Since ∑iui=0\sum_i u_i=0, summing (25) yields

W(u)≥b−a2∥u∥1−∑iCi,u∈H.W(u)\ge\frac{b-a}{2}\lVert u\rVert_1-\sum_i C_i,\qquad u\in H.

Thus WW is coercive on the closed subspace HH and admits a minimizer uˉ∈H\bar{u}\in H. The first-order condition is ∇W(uˉ)∈H⊥=span⁡{1N}\nabla W(\bar{u})\in H^\perp=\operatorname{span}\{\mathbf{1}_N\}, that is, sft⁡i(uˉi)=y∗\operatorname{sft}_i(\bar{u}_i)=y^* for every ii, for some y∗∈Ry^*\in\mathbb{R}. Setting ηˉ=−uˉ\bar{\eta}=-\bar{u} supplies the required zero-sum configuration. Notice that this coercivity estimate is independent of strict monotonicity and of whether the shift maps have finite saturation bounds. In particular, the same argument applies to the Lyapunov potential V(η)=W(−η)V(\eta)=W(-\eta) appearing in Theorem 1, with η∈im⁡(L)\eta\in\operatorname{im}(L). The corresponding minimizer set is therefore nonempty and compact, even when different input values yield the same common velocity. This observation separates the existence of a compatible input vector from the uniqueness of its associated output value.

For uniqueness of the compatible velocity, Suppose y1<y2y_1<y_2 both admitted such a configuration, and let ui(r)∈sft⁡i−1(yr)u_i^{(r)}\in\operatorname{sft}_i^{-1}(y_r) with ∑iui(r)=0\sum_i u_i^{(r)}=0, r=1,2r=1,2. Because each sft⁡i\operatorname{sft}_i is non-decreasing, every preimage of y1y_1 is strictly smaller than every preimage of y2y_2. Consequently ui(1)<ui(2)u_i^{(1)}<u_i^{(2)} for all ii, contradicting both zero-sum equalities. Hence y∗y^* is unique. □

The geometric interpretation in Figure 10 uses the preimage intervals Ti(y):={ηi:sft⁡i(−ηi)=y}T_i(y):=\{\eta_i:\operatorname{sft}_i(-\eta_i)=y\} and their Minkowski sum T(y):=∑iTi(y)T(y):=\sum_i T_i(y) for y∈Ry\in\mathbb{R}. Each Ti(y)T_i(y) is an interval because sft⁡i\operatorname{sft}_i is continuous and non-decreasing. For y1<y2y_1<y_2, every element of Ti(y1)T_i(y_1) exceeds every element of Ti(y2)T_i(y_2); hence the admissible sums move in the opposite direction as the common velocity increases. The proof of Lemma 2 establishes that precisely one y∗∈Ry^*\in\mathbb{R} admits 0∈T(y∗)0\in T(y^*), without assuming bounded inverses or continuity of preimage endpoints at saturation. Figure 10 illustrates this compatibility relation rather than supplying the existence argument. The left panel shows the two-agent case; the right panel depicts the corresponding interval sums as the candidate common velocity varies. Flat intervals in the shifted maps can make the compatible vector ηˉ\bar{\eta} non-unique, even though its associated velocity y∗y^* is unique.

Figure 10. Illustration used in Lemma 2. The left subfigure shows two shifted functions, evaluated at −ηi-\eta_i, and the aligned velocity value y∗y^* for which the corresponding variables ηi\eta_i, i=1,2i=1,2, sum to zero. The right subfigure shows the possible values of ∑i∈Vηi\sum_{i\in\mathcal{V}}\eta_i for each aligned velocity y=sft⁡i(−ηi)y=\operatorname{sft}_i(-\eta_i) and their monotone variation with respect to yy.

Flat actuator regions may admit several zero-sum compatible input vectors at the same velocity y∗y^*. Theorem 1 nevertheless guarantees convergence of each trajectory to one such vector, potentially depending on its initial condition. Under global invertibility the compatible vector is unique and is determined by the shift maps.

Proof of Lemma 3

Proof: Let ∇f(x)=col⁡(h1(x1),…,hk(xk))\nabla f(x)=\operatorname{col}(h_1(x_1),\ldots,h_k(x_k)) with each hih_i non-decreasing, and fix x,y∈Rkx,y\in\mathbb{R}^k. Integrating ddtf(x+t(y−x))\frac{\mathrm{d}}{\mathrm{d}t}f(x+t(y-x)) over [0,1][0,1] gives f(y)−f(x)−∇f(x)⊤(y−x)=∫01∑i=1k(hi(xi+t(yi−xi))−hi(xi))(yi−xi) dt≥0f(y)-f(x)-\nabla f(x)^\top(y-x)=\int_0^1\sum_{i=1}^k\bigl(h_i(x_i+t(y_i-x_i))-h_i(x_i)\bigr)(y_i-x_i)\,\mathrm{d}t\ge0, since each summand is non-negative by monotonicity of hih_i. This is the first-order characterization of convexity. □

Proof of Lemma 4

Proof: Since ff is convex and differentiable, any stationary point is a global minimizer; hence X∗=arg⁡min⁡f\mathcal{X}^*=\arg\min f, and we let f∗:=f(ξ∗)f^*:=f(\xi^*) for ξ∗∈X∗\xi^*\in\mathcal{X}^*.

To establish radial unboundedness, argue by contradiction. Assume ff is not radially unbounded. Then there exists some c>f∗c>f^* such that the sublevel set Λc:={ξ∈Rk:f(ξ)≤c}\Lambda_c:=\{\xi\in\mathbb{R}^k:f(\xi)\le c\} is unbounded. The set Λc\Lambda_c is closed and convex, and every unbounded closed convex set contains a ray. Indeed, fix α∈Λc\alpha\in\Lambda_c and take ξn∈Λc\xi_n\in\Lambda_c with ∥ξn∥→∞\lVert\xi_n\rVert\to\infty. Up to a subsequence, (ξn−α)/∥ξn−α∥→v(\xi_n-\alpha)/\|\xi_n-\alpha\| \to v with ∥v∥=1\|v\|=1. For each t≥0t \ge0 and nn large, α+t(ξn−α)/∥ξn−α∥∈Λc\alpha+ t(\xi_n-\alpha)/\|\xi_n-\alpha\| \in\Lambda^c by convexity, so α+tv∈Λc\alpha+ tv \in\Lambda^c by closedness. Now fix ξ⋆∈X⋆\xi^\star\in X^\star, t≥0t \ge0, and λ∈(0,1]\lambda\in(0,1]. By convexity,

f((1−λ)ξ⋆+λ(α+tλv))≤(1−λ)f⋆+λc.f\left((1-\lambda)\xi^\star+\lambda\left(\alpha+\frac{t}{\lambda}v\right)\right)\leq(1-\lambda)f^\star+\lambda c.

The argument on the left equals (1−λ)ξ⋆+λα+tv(1-\lambda)\xi^\star+\lambda\alpha+tv, which tends to ξ⋆+tv\xi^\star+tv as λ→0\lambda\to0, while the right-hand side tends to f⋆f^\star. By continuity of ff, f(ξ⋆+tv)≤f⋆f(\xi^\star+tv)\leq f^\star, so ξ⋆+tv∈X⋆\xi^\star+tv \in X^\star for every t≥0t \geq0. This contradicts the boundedness of X⋆X^\star. Therefore, every sublevel set of ff is bounded, which is equivalent to f(ξ)→+∞f(\xi)\to+\infty as ∥ξ∥→∞\|\xi\|\to\infty. In other words, ff is radially unbounded. □

References

  1. [1]F. Bullo, J. Cortés, F. Dörfler, and S. Martínez, Lectures on network systems. CreateSpace PlatformScotts Valley, 2018, vol. 1.
  2. [2]Y. Xu, J. Bautista, J. Hinojosa, and H. Garcia de Marina, “Distributed oscillatory guidance for formation flight of fixed-wing drones,” in 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 2584–2591.
  3. [3]H. Zhang and P. Gurfirl, “Satellite cluster flight using on-off cyclic control,” Acta Astronautica, vol. 106, pp. 1–12, 2015.DOI
  4. [4]T. Yang, Z. Meng, D. V. Dimarogonas, and K. H. Johansson, “Global consensus for discrete-time multi-agent systems with input saturation constraints,” Automatica, vol. 50, no. 2, pp. 499–506, 2014.DOI
  5. [5]Y.-H. Lim and H.-S. Ahn, “Consensus under saturation constraints in interconnection states,” IEEE Transactions on Automatic Control, vol. 60, no. 11, pp. 3053–3058, 2015.DOI
  6. [6]Y. Zou, Z. Zuo, X. Xia, and Z. Ding, “Consensus of saturated multiagent systems: Tolerance to nonidentical asymmetric saturation levels,” IEEE Transactions on Automatic Control, vol. 69, no. 3, pp. 1804–1811, 2024.DOI
  7. [7]G. Wang and Z. Zuo, “Consensus algorithms for double-integrator dynamics with different velocity and actuator saturation constraints,” Automatica, vol. 170, p. 111860, 2024.DOI
  8. [8]H. Zhu, J. Lu, Y. Lou, and X. Yang, “Distributed saturated impulsive quasi-consensus for leader-follower multi-agent systems: An open topology framework,” IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 9, pp. 1941–1943, 2025.DOI
  9. [9]J. Fu, G. Wen, and X. Yu, “Safe consensus tracking with guaranteed full state and input constraints: A control barrier function-based approach,” IEEE Transactions on Automatic Control, vol. 68, no. 12, pp. 8075–8081, 2023.DOI
  10. [10]P. Lin, Y. Wang, D. Wang, and Z.-G. Wu, “Constrained containment control for multiagent systems with possible empty intersection of position constraints,” IEEE Transactions on Automatic Control, vol. 69, no. 3, pp. 1842–1849, 2024.DOI
  11. [11]J. Wang, S. Li, and Y. Zou, “Connectivity-maintaining consensus of multi-agent systems with communication management based on predictive control strategy,” IEEE/CAA Journal of Automatica Sinica, vol. 10, no. 3, pp. 700–710, 2023.DOI
  12. [12]P. Sinha, S. Srikant, and H. Sinhmar, “Consensus of networked double integrator systems under sensor bias,” International Journal of Adaptive Control and Signal Processing, vol. 37, no. 1, pp. 278–297, 2023.arxiv.org/abs/2204.08666
  13. [13]H. Chen, D. Zelazo, X. Wang, and L. Shen, “Convergence analysis of signed nonlinear networks,” IEEE Transactions on Control of Network Systems, vol. 7, no. 1, pp. 189–200, 2019.arxiv.org/abs/1810.11274
  14. [14]H. Garcia de Marina, M. Cao, and B. Jayawardhana, “Controlling rigid formations of mobile agents under inconsistent measurements,” IEEE Transactions on Robotics, vol. 31, no. 1, pp. 31–39, 2015.DOI
  15. [15]Z. Sun, S. Mou, B. D. O. Anderson, and A. S. Morse, “Formation movements in minimally rigid formation control with mismatched mutual distances,” in Proceedings of the 53rd IEEE Conference on Decision and Control. IEEE, 2014, pp. 6161–6166.DOI
  16. [16]H. Garcia de Marina, B. Jayawardhana, and M. Cao, “Distributed rotational and translational maneuvering of rigid formations and their applications,” IEEE Transactions on Robotics, vol. 32, no. 3, pp. 684–697, 2016.arxiv.org/abs/1604.07849
  17. [17]H. G. de Marina, “Maneuvering and robustness issues in undirected displacement-consensus-based formation control,” IEEE Transactions on Automatic Control, vol. 66, no. 7, pp. 3370–3377, 2021.arxiv.org/abs/2008.03544
  18. [18]H. K. Khalil and J. W. Grizzle, Nonlinear systems. Prentice hall Upper Saddle River, NJ, 2002, vol. 3.
  19. [19]L. C. Evans, Partial differential equations. American mathematical society, 2022, vol. 19.
  20. [20]J. C. Strikwerda, Finite difference schemes and partial differential equations, Second Edition. Society for Industrial and Applied Mathematics, 2004.DOI
  21. [21]S. Motsch and E. Tadmor, “Heterophilious dynamics enhances consensus,” SIAM Review, vol. 56, no. 4, pp. 577–621, 2014.
  22. [22]F. Vasca, C. Bernardo, and R. Iervolino, “Practical consensus in bounded confidence opinion dynamics,” Automatica, vol. 129, p. 109683, 2021.DOI
  23. [23]J. M. Hendrickx and J. N. Tsitsiklis, “Convergence of type-symmetric and cut-balanced consensus seeking systems,” IEEE Transactions on Automatic Control, vol. 58, no. 1, pp. 214–218, 2013.DOI
  24. [24]W. Ren and R. Beard, “Consensus seeking in multiagent systems under dynamically changing interaction topologies,” IEEE Transactions on Automatic Control, vol. 50, no. 5, pp. 655–661, 2005.

Paper details

vlnc:W_ujevu3cn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents