WorksComputer Science
End-to-End Reinforcement Learning for Quadrupedal Jumping in Lunar Gravity
Abstract
Legged robots can extend planetary exploration to rugged and discontinuous terrain that is inaccessible to conventional rovers. In lunar gravity, jumping further expands this mobility by enabling robots to cross obstacles and gaps that cannot be traversed by walking alone. We present a multistage reinforcement-learning approach that produces a single end-to-end control policy for commanded point-to-point jumping under lunar gravity. The resulting policy has a simple architecture and relies exclusively on mission-compatible onboard sensing, making its deployment on a real lunar mission plausible. In simulation, the learned policy performs accurate point-to-point jumps with landing errors smaller than the robot's body size. We analyze the torque and power demands of lunar jumping and characterize performance degradation under extreme outlier events, including control-update freezes of up to \SI{1}{s} during flight. Hardware transfer demonstrates that the policy can successfully execute a vertical jump under lunar-gravity offloading. Furthermore, the measured torque and power profiles are consistent with the simulation results and remain within the robot's hardware limits.
INTRODUCTION
Robotic exploration of the solar system is essential for advancing scientific understanding of planetary bodies and preparing future human missions. While orbiters and stationary landers provide valuable measurements, scientifically relevant targets require mobile surface systems to acquire local observations and interact with the terrain. As interest grows in sustained surface operations [1], infrastructure construction [2], [3], and scientific access to challenging terrain, missions increasingly target sites such as permanently shadowed regions, crater walls, and lunar lava tubes – areas that remain difficult to reach with conventional mobility systems [4, 5, 6, 7, 8, 9, 10]. Planetary rovers have so far relied predominantly on wheeled locomotion, which provides substantial flight heritage and reliability on comparatively benign terrain, but remains limited on steep, loose, rugged, or discontinuous surfaces [11]. In contrast, recent advances in deep reinforcement learning have substantially improved the ability of legged robots to traverse rough and unstructured terrain, motivating their investigation for planetary exploration.
Beyond robust walking, the morphology of legged robots also enables dynamic locomotion modes such as jumping [12]. For a lunar robot, jumping is not intended to replace walking as an energy-efficient mode of long-range traversal. Still, it can serve as a useful mobility primitive for overcoming obstacles, terrain discontinuities, slopes, and gaps that would otherwise be untraversable [13]. In low gravity,

Figure 1. The Magnecko robot mid-flight under gravity offloading during hardware deployment of the jumping controller.
such maneuvers are especially attractive because relatively moderate launch velocities can produce extended ballistic flight phases [12].
At the same time, lunar jumping imposes additional control and deployment challenges: the controller must demonstrate high robustness, particularly during flight and landing, to avoid crashing onto the surface. Specifically on non-redundant system architectures, FDIR may even require the ability to recover from a mid-flight system reboot. In addition, constraints in computing hardware for aerospace engineering favor the development of a low-complexity controller. Although quadrupedal jumping has been demonstrated in both terrestrial and low-gravity contexts, commanded point-to-point jumping for a quadruped in lunar gravity has not yet been sufficiently investigated.
To address this gap, we develop a deployment-lean reinforcement-learning framework for commanded point-to-point jumping in lunar gravity. The key idea is to pair the governing ballistic properties of lunar flight with a multistage control scheme. This provides a versatile training framework to synthesize an end-to-end deep reinforcement learning-based jumping controller within a single multilayer perceptron control policy. The control policy maps onboard-feasible observations to joint-position targets.
Specifically, our contributions are:
We formulate commanded point-to-point jumping in lunar gravity by exploiting the ballistic flight structure to relate the target position and flight trajectory-apex height to the desired launch velocity.
We introduce an end-to-end RL framework that learns the full jumping maneuver using a single deployable control policy, while avoiding dense motion references,
online trajectory generation, and policy switching.
We evaluate the resulting controller on the Magnecko [14] quadruped for jumping accuracy, recovery, and robustness to random failures.
We transfer the resulting controller to gravity offloading hardware to evaluate the sim-to-real profile of the torque and power curves.
RELATED WORK
Quadrupedal jumping has been achieved with trajectory optimization and online MPC tracking [15, 16, 17], and with fast online trajectory generation [18, 19]. Chignoli et al. additionally use an offline jump-feasibility classifier to select omnidirectional jumps in real time [20]. These architectures are highly complex and depend on substantial onboard computation and accurate dynamics models, which makes them ill-suited to a space-deployment setting.
Reinforcement learning (RL), by contrast, is generally more robust and cheaper at rollout time. A common strategy shapes training with a motion reference: Bellegarda et al. track optimized trajectories with a learned policy [21], Liu et al. learn the parameters of a hand-designed foot trajectory for distance-controllable jumps [22], and others imitate optimized or example trajectories [23, 24, 25]. In low gravity, Sang and Wang combine a main-point trajectory generator with staged rewards and shared experience across tasks [26], which Ye et al. later augment with behavior cloning to reach higher relief [27]. While motion references improve robustness, they make controller development tedious and limit the jump to the performance of the reference; moreover, reference-based jumping has, to our knowledge, not been transferred to low-gravity or offloaded hardware.
Other works avoid references entirely. Atanassov et al. obtain a single goal-conditioned policy spanning a wide range of jumps purely through curriculum design, though only under Earth gravity and dense velocity guidance [28].
Hierarchical or switching controllers that pair a learned high-level policy with model-based or per-phase low-level controllers add capability but introduce additional modules and potential points of failure at deployment [29, 30]. Closest to our setting, Xu et al. demonstrate continuous lunar pronking on a hardware-in-the-loop digital-twin testbed that pairs an offload with a tilting platform and treadmill [31]. While the latter shows promising results, continuous pronking is undesirable in a mission scenario due to the high cost of transport and unnecessary risk. We instead focus on a single jump, to be deployed only when necessary.
METHOD
We train and test our controller on an improved version of Magnecko [14]. The robot weighs 15.6 kg, has a square base with side length 0.25 m, and features 3 actuated joints per leg for hip abduction/adduction (HAA), hip flexion/extension (HFE), and knee flexion/extension (KFE). In the default configuration with extended legs, the robot’s total extent is a square with side length 0.62 m.
We emphasize the deployability of our controller in space. For this reason, we deliberately avoid using secondary estimation modules and recurrent structures in our controller, which could be points of failure in the event of temporary loss of compute.
Jumping Physics and Task Formulation
Figure 2 illustrates the jumping flight phases over time. Due to the lack of an atmosphere on the moon, once the robot has disconnected from the ground, the trajectory of the robot’s center of mass is fully determined by the closed-form equations of ballistic flight. A variety of parabolas with varying apex heights connect any two points on the surface. To simplify the learning problem, we uniquely determine the desired parabola path in our command space. The policy receives a command

Figure 2. Illustration of the five-stage control scheme used during training and its transition logic. The control policy does not observe the current stage.
where is the touchdown position relative to the starting position of the robot under a flat ground assumption, and is the apex height of the parabola. This allows the user to choose the parabola’s shape to suit the desired use case. The actor’s observations are
where are joint positions, are joint velocities, are previous actions, is projected gravity, is time since beginning of the jumping maneuver, and are the last 8 IMU angular velocity readings. To aid training, the critic additionally receives privileged information, namely the current stage, the foot contact states, the world pose of the robot base, , the robot base’s twist, the world position of the robot’s center of mass, the world velocity of the robot’s center of mass, and the optimal launch velocity given the parabola derived from the command.
The actions control the proportional-derivative (PD) controllers on the robot’s joints and are updated at . The robot’s joint position targets are computed as
where is the action scaling factor, and are the default joint positions.
At each episode, a jump command is sampled according to the ranges in Table I. The corresponding launch velocity, an alternative to uniquely define the commanded jumping parabola which we use to reformulate the sparse point-to-point jumping to a launch velocity tracking problem, is reconstructed analytically as
| Category | Parameter | Value |
| Command Sampling | Jump distance | |
| Command Sampling | Heading | |
| Command Sampling | Launch elevation | |
| Command Sampling | Launch speed | |
| Command Sampling | Apex height | |
| Terrain | Flat terrain friction | |
| Terrain | Slope inclination | |
| Terrain | Slope azimuth | |
| Randomization | Base mass | |
| Randomization | Static friction | |
| Randomization | Dynamic friction | |
| Curriculum | Initial elevation lower bound | |
| Curriculum | Final elevation lower bound | |
| Curriculum | Annealing schedule | every 500 iterations after iteration 1000 |
| Regularization | Actuator-power weight | linearly ramped from 0 |
| Regularization | Ramp duration | first 1000 iterations |
Table I. Simulation and training parameters.
Policies are trained on flat terrain and planar-sloped patches, with terrain parameters, domain randomization, and curriculum settings summarized in Table I. Symmetry augmentation is applied over the symmetry axes of the robot, which improves convergence and base-attitude stability without additional simulation [32].
Multistage Scheme
We decompose point-to-point jumping into five distinct stages: STANCE (0), LAUNCH (1), FLIGHT (2), LANDING (3), and RECOVERY (4). All stages share the same regularizations, and each stage has minimal additional rewards to encourage desirable behavior. The active stage is determined by the transition logic shown in Figure 2 and Table II. The full reward breakdown and the stage in which each term is active are given in Table III. The current stage is not part of the policy’s observations.
| Transition | Condition | Param. | Value |
| STANCELAUNCH | 2 s | ||
| LAUNCHFLIGHT | 0.75 m | ||
| FLIGHTLANDING | 0.75 m | ||
| LANDINGRECOVERY | 2 s |
Table II. Stage transition conditions and parameters. is set once at least three feet are simultaneously in contact after takeoff.
| Term | Expression | Stage (weight) | Parameters |
| Task rewards & penalties | |||
| Start Base Pos. | STA 0.5 | ||
| Start Base Head. | STA 0.5 | ||
| Launch Speed | , (5) (8) | LAU 25.0 | |
| Launch Direction | , (6) (8) | LAU 15.0 | |
| Bad Launch | (10) (9) | LAU −10.0 | m/s, |
| Flat Base | FLI, REC 0.5 | ||
| Feet in Contact | STA 2.0; LAN, REC 0.5 | — | |
| Recovery Height | REC 0.2 | m | |
| Regularizations | |||
| Actuator Power | ALL −5×10⁻⁵ | — | |
| Joint Limits | ALL −2.0 | — | |
| Base Contact | ALL −100.0 | N | |
| Leg Contact | ALL −20.0 | N |
Table III. Reward terms, expressions, active stages with weights, and parameters.
The STANCE stage gives the robot time to prepare for the jump. We give a reward to prevent deviation from initial position during this stage, but do not impose any type of ’wind-up’ reference. After , the stage is transitioned to LAUNCH.
The LAUNCH stage is the most important part of the maneuver. During this stage, the robot receives separate rewards for tracking the launch speed and direction. Both share a convex combination of a wide () and a narrow () kernel: the wide kernel ensures sufficient signal during early training. In contrast, the narrow kernel enforces the precision required once the basic behavior has been discovered.
The Launch Speed reward applies a Gaussian kernel to the relative center of mass (COM) speed error
where using the relative error makes the term independent of the commanded jump magnitude. We apply the reward twice, once with a large to encourage discovery of the reward, and once with a smaller to encourage precision.
The Launch Direction reward,
applies an exponential kernel to the cosine alignment error,
where is the angle between the executed and commanded COM launch direction. Using the cosine avoids the noise of explicitly computing the angular error. As with the speed reward, we apply it twice, once with a large and once with a smaller .
TABLE III: Reward terms, expressions, active stages with weights, and parameters. Stages: STA = STANCE, LAU = LAUNCH, FLI = FLIGHT, LAN = LANDING, REC = RECOVERY; “ALL” denotes a term active in every stage. The LAUNCH task rewards and , modulated by the temporal kernel , are defined in (5)–(8).
We further emphasize a single, strong jump by applying a temporal exponential kernel
over the task rewards , which rewards tracking the required launch velocity instantaneously rather than finding a local minimum, in which the policy makes multiple small hops. Using this trick, the policy learns to perform a ’wind-up’ motion at the end of the STANCE stage. The robot transitions from the LAUNCH stage to FLIGHT once a minimum height is reached. We use two ’bad launch’ termination conditions. The first is triggered if the robot has failed to launch completely after a set time
where s for this work. The second termination condition is triggered when the robot successfully launches, but does so with unacceptable launch parameters:
where are launch velocity and direction tracking error respectively. The robot also incurs a large penalty when terminated under either of these conditions. This both accelerates and stabilizes the learning, as the robot spends less time in junk states where it failed to launch properly. Importantly, this training setup causes the robot to launch immediately at with . Notably, due to the precisely timed and executed launch velocity and as we provide a timer observation, the robot can infer its progress on the flight trajectory, without dedicated exteroceptive sensing information. In fact, in conjunction with the asymmetric actor-critic setup, and the type of stage transition heuristics in Table II on which the set of active rewards ultimately depends, this not only allows us to synthesize the distinct stage-specific control behavior within a single control policy but to do so without relying on dense motion references, neither during training nor deployment, and with minimal mission-compatible sensing requirements.
During the underactuated FLIGHT stage, the Flat Base reward incentivizes the robot to leverage conservation of angular momentum and coordinated limb motion for in-flight base tilt control, resp. to keep the base stable. The stage transitions to LANDING when the robot’s base falls below the threshold . During the LANDING stage, the robot is rewarded for maximizing the number of feet in contact with the ground, to encourage spreading out the impact as much as possible. As soon as at least 3 feet are in contact with the ground, the stage transitions to the terminal RECOVERY stage. In this stage, the robot is rewarded for staying put and adopting a stable stance. This allows for easy transition back into a ’standard’ locomotion control.
Simulation and Hardware Setup
We train our control policy in IsaacLab [33] with the Proximal Policy Optimization (PPO) algorithm [34] and the hyperparameters from Rudin et al. [35]. We use multilayer perceptrons with hidden layer sizes of and ELU activation functions for both the actor and critic networks. The simulation joint parameters of the robot were identified using PACE [36]. The control policies were trained for 10,000 apex height, on an RTX4090 GPU.
For hardware deployment, Magnecko was attached to a custom gravity offloading mechanism that approximates lunar-gravity loading. The mechanism uses a constant-force spring reel to provide an upward tether force that offsets approximately 5/6 of the robot’s gravitational load, leaving an effective weight close to that under lunar gravity. The legs are provided feedforward gravity offloading torques as in Arm et al. [37] such that the legs experience the same gravity as in the simulation. Although the setup does not reproduce fully unconstrained lunar flight, it provides sufficient fidelity and a practical hardware testbed for vertical jumping experiments and for comparing simulation and hardware results.
RESULTS
Jumping Accuracy
We evaluated the landing position accuracy over the trained workspace for both flat and sloped terrain. We randomly sampled 10,000 triplets of landing position , and landing position and apex height. These commands were then rolled out once on flat terrain and once on sloped terrain. The sloped terrain’s incline was sampled from and its direction was sampled from . The landing accuracy is shown in Figure 3. In both scenarios, tracking accuracy appears to be better when the commanded apex is larger and landing accuracy degrades with commanded jumping distance. The degradation is more pronounced for the sloped scenario.

Figure 3. Landing accuracies corresponding to 10,000 commands covering the full command workspace. Top: Flat Terrain. Bottom: Sloped Terrain.
Robustness
We investigated the robustness of the trained control policy in two experiments. In the first experiment, we initialized the control policy mid-flight in a random orientation, as one might encounter, for example, as a result of a single-event upset and consequent forced reboot of the controlling computer. The starting configuration roll, pitch, and yaw angles were sampled from , and starting angular velocities were sampled from . We then measured the time required for the control policy to reorient the robot base to its nominal tilt. The convergence threshold chosen for this was . The median convergence time was 0.7 s, with 99 % converging in under 1.5 s. A representative mid-air recovery sequence is shown in Figure 4.

Figure 4. Mid-air recovery behavior of the controller.
In the second experiment, we ran 10,000 rollouts on flat and sloped terrain using the same randomly sampled commands as in § 4-A. We then logged contact forces between the robot base and the terrain as a proxy for landing robustness. Any contact with the robot’s base and the ground was considered as a crash regardless of contact force magnitude. On flat terrain, 92.5 % of robots completed the jump without any contact forces. For the robots jumping on sloped terrain, 88.1 % completed the jump without any contact forces. We then ran the experiment with simulated failures. For each rollout, we briefly disabled the control policy’s target inputs to the joints at a random time during the episode, simulating a compute failure. Upon regaining control, we flushed the IMU angular velocity history buffer to simulate loss of previous readings during compute shutdown. We performed the experiment three times, simulating loss of control for seconds. The joints themselves remained active with the last sent PD target, as we assume they have an inbuilt driver. Table IV shows the failure rates.
| Failure duration | Flat Crashes | Sloped Crashes |
| 0.0 s (nominal) | 7.5% | 11.9% |
| 0.2 s | 7.8% | 12.5% |
| 0.5 s | 11.3% | 15.6% |
| 1.0 s | 16.7% | 20.3% |
Table IV. Failure rates under fixed-duration control freeze.
Torque and Power Demands
To characterize the mechanical and electrical demands imposed by jumping, both joint-level torques and actuator input power demands were analyzed. In this regard, the takeoff sequence is the dominant event, with pronounced torque and power peaks immediately before takeoff is recorded. In a separate experiment on sloped terrain, we summarized the mean and envelope of the torques for each joint group, and the combined input power demands over all actuators around takeoff, indicated by the black triangles in Figure 5 for a command set that systematically covers the entire trained command domain. The Magnecko actuators have a peak torque of 56 Nm.

Figure 5. Simulated torque and power demands of the robot around takeoff; dashed line and black triangle mark takeoff time.
Hardware Feasibility
We investigated the sim-to-real transfer behavior. The robot was attached to the gravity offloading device de-
scribed in [37]. As the offload was incapable of translational movements, we commanded purely vertical jumps with an apex height of 0.8 m, and and positional commands of 0 m, effectively jumping in place. The height of 0.8 m was chosen due to the limits of the gravity offload. We focus on takeoff, as it is the most critical part of the maneuver, since the trajectory is fully determined at takeoff in the absence of an atmosphere. Figure 6 compares the torque distributions measured in the hardware experiments against a matched simulation rollout. It also reports the total actuation hardware power usage and the corresponding simulation estimate around takeoff. Figure 7 demonstrates the hardware jumping sequence.

Figure 6. Sim-to-real joint value transfer. Left: per joint group torque distributions during the launch for simulation and hardware deployment. Right: total actuator input power demand during the launch for simulation and hardware deployment.

Figure 7. Example of the jumping sequence deployed on Magnecko.
DISCUSSION
For launch trajectories with elevation angles , the landing error generally remains below 0.3 m, less than half of the robot’s 0.62 m total extent. This indicates that the controller provides sufficient landing accuracy for its intended role as a contingency maneuver to overcome otherwise untraversable terrain, where exact touchdown placement is less critical than safely reaching the target region. The reduced accuracy on sloped terrain is consistent with the flat-ground assumption used to derive the commanded ballistic trajectory. This limitation can be mitigated during mission operation by adjusting the commanded touchdown position and apex height, either manually by an operator or automatically by a higher-level planner.
The robustness experiments demonstrate that the proposed controller can quickly and consistently reorient the robot in mid-air. In nominal rollouts, the policy achieved collision-free landings in of trials on both flat and sloped terrain. Importantly, the controller remained highly resilient even under severe compute interruptions, such as a random control freeze lasting up to 1 s, and the robot still survived 79.7% of sloped-terrain rollouts. This robustness to temporary loss of control authority is encouraging for deployment-constrained systems, where transient faults may occur. Although mission use would require appropriate risk assessment, the results support jumping as a credible contingency behavior for escaping otherwise untraversable terrain.
The torque and power evaluation confirms the running hypothesis that actuation requirements are mostly determined by the takeoff sequence, which imposes demanding short-time torques and powers, provides a broad reference for system design, and shows that even for the most demanding jumps, demands respect Magnecko’s hardware constraints.
The hardware deployment shows that the sim-to-real gap can be bridged for vertical real-world jumping. Horizontal jumping could not be validated due to limitations in the gravity offload.
CONCLUSION AND FUTURE WORK
We have introduced a method for robust point-to-point jumping for a legged robot in lunar gravity. The method is deliberately simple in architecture to facilitate deployment in space applications where secondary estimation modules and recurrence may introduce deployment risk. We have shown that jumping is feasible to implement on hardware and robust enough to deploy as a last resort should one encounter terrain that would be otherwise untraversable. While the validation done with the gravity offload provides a promising hardware transfer, due to limitations in the offload’s design we were unable to validate horizontal jumps. The most immediate followup to this work should be developing and validating on an improved gravity offloading device which displays low inertia in the horizontal direction to conclusively prove hardware transfer fully possible. Additionally, the method’s robustness and success rate could be further improved to make it useful beyond deployment as a last resort. Future work should aim to improve robustness while retaining a simple architecture. Finally, the ground contact is currently modeled as rigid, whereas the lunar regolith provides a soft contact that should be properly modeled in future work.
ACKNOWLEDGMENTS
We thank the Magnecko team for their continued support with repairing the robot, the maintenance of the robot’s software and hardware, and, more generally, for building a well-engineered and easy-to-use system.
LunarLeaper acknowledges funding through the CHiwi Foundation, SERI’s NASO program (1-012571, LunarLeaper 2024), Moog Inc., maxon, and MARVIS.
References
- [1]A. Krawcic, et al., “Lunar rover cargo transport: Mission concept and field test,” IEEE Transactions on Field Robotics, 2026.
- [2]K. Uno, et al., “Moonbot: Modular and on-demand reconfigurable robot towards moon base construction,” IEEE Transactions on Field Robotics, 2025.
- [3]W. Brinkmann, et al., “Enhancement of the six-legged robot mantis for assembly and construction tasks in lunar mission scenarios within a multi-robot team,” in Proceedings: International Symposium on Artificial Intelligence, Robotics and Automation in Space, 2020.
- [4]R. Jangale, et al., “Novel robotic fleet for sample recovery in lunar craters: A concept of operations,” IEEE Transactions on Field Robotics, vol. 2, pp. 937–955, 2025.DOI
- [5]F. Cordes, et al., ““Lunares: Lunar crater exploration with heterogeneous multi robot systems,” Intelligent Service Robotics, vol. 4, no. 1, pp. 61–89, 2011.DOI
- [6]H. Kolvenbach, et al., “Lunarleaper—a mission concept to explore the lunar subsurface with a small-scale legged robot,” Acta Astronautica, 2025.
- [7]A. Parness, et al., “LEMUR 3: A limbed climbing robot for extreme terrain mobility in space,” in 2017 IEEE International Conference on Robotics and Automation (ICRA), May 2017, pp. 5467–5473.DOI
- [8]Y. Tanaka, et al., “Scaler: Versatile multi-limbed robot for free-climbing in extreme terrains,” IEEE Transactions on Robotics, 2025.
- [9]A. Schperberg, et al., “Mobius: A multi-modal bipedal robot that can walk, crawl, climb, and roll,” Robotics: Science and Systems, 2026.arxiv.org/abs/2511.01774
- [10]K. Uno, et al., “Limbero: A limbed climbing exploration robot toward traveling on rocky cliffs,” arXiv preprint arXiv:2603.16531, 2026.arxiv.org/abs/2603.16531
- [11]G. Ishigami, et al., “Slope traversal controls for planetary exploration rover on sandy terrain,” Journal of Field Robotics, vol. 26, no. 3, pp. 264–286, 2009.DOI
- [12]A. Spiridonov, et al., “Spacehopper: A small-scale legged robot for exploring low-gravity celestial bodies,” arXiv preprint arXiv:2403.02831, 2024.arxiv.org/abs/2403.02831
- [13]P. Arm, et al., “Spacebok: A dynamic legged robot for space exploration,” in 2019 International Conference on Robotics and Automation (ICRA), 2019, pp. 6288–6294.DOI
- [14]S. Leuthard, et al., “Magnecko: Design and control of a quadrupedal magnetic climbing robot,” in Climbing and Walking Robots Conference. Springer, 2024, pp. 55–67.DOI
- [15]C. Nguyen and Q. Nguyen, “Contact-timing and trajectory optimization for 3d jumping on quadruped robots,” in 2022 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE, 2022, pp. 11 994–11 999.arxiv.org/abs/2110.06764
- [16]Y. Tanaka, et al., “Tethered variable inertial attitude control mechanisms through a modular jumping limbed robot,” in 2025 IEEE Aerospace Conference. IEEE, 2025, pp. 1–11.arxiv.org/abs/2501.10156
- [17]C. Nguyen, et al., “Continuous jumping for legged robots on stepping stones via trajectory optimization and model predictive control,” in 2022 IEEE 61st Conference on Decision and Control (CDC). IEEE, 2022, pp. 93–99.arxiv.org/abs/2204.01147
- [18]L. Yue, et al., “A fast online omnidirectional quadrupedal jumping framework via virtual-model control and minimum jerk trajectory generation,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 11 993–11 999.arxiv.org/abs/2407.00658
- [19]G. Bellegarda, et al., “Quadruped-frog: Rapid online optimization of continuous quadruped jumping,” in 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 1443–1450.arxiv.org/abs/2403.06954
- [20]M. Chignoli, et al., “Rapid and reliable quadruped motion planning with omnidirectional jumping,” in 2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 6621–6627.arxiv.org/abs/2111.13648
- [21]G. Bellegarda, et al., “Robust quadruped jumping via deep reinforcement learning,” Robotics and Autonomous Systems, vol. 182, p. 104799, 2024.arxiv.org/abs/2011.07089
- [22]Q. Liu, et al., “Distance-controllable long jump of quadruped robot based on parameter optimization using deep reinforcement learning,” IEEE Access, vol. 11, pp. 98 566–98 577, 2023.DOI
- [23]J. Ding, et al., “Robust jumping with an articulated soft quadruped via trajectory optimization and iterative learning,” IEEE Robotics and Automation Letters, vol. 9, no. 1, pp. 255–262, 2023.DOI
- [24]G. Apostolidis, et al., “Explosive jumping with rigid and articulated soft quadrupeds via example guided reinforcement learning,” arXiv preprint arXiv:2503.16197, 2025.
- [25]R. Wang, et al., “Integrating trajectory optimization and reinforcement learning for quadrupedal jumping with terrain-adaptive landing,” in 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 18 926–18 933.arxiv.org/abs/2509.12776
- [26]H. Sang and S. Wang, “Lunar leap robot: 3m architecture–enhanced deep reinforcement learning method for quadruped robot jumping in low-gravity environment,” Journal of Aerospace Engineering, vol. 37, no. 6, p. 02404706, 2024.DOI
- [27]Y. Ye, et al., “Demonstration augmented-drl: An architecture for quadruped robot jumping on lunar high-relief terrain,” in 2026 12th International Conference on Electrical Engineering, Control and Robotics (EECR), 2026, pp. 200–207.DOI
- [28]V. Atanassov, et al., “Curriculum-based reinforcement learning for quadrupedal jumping: A reference-free design,” IEEE Robotics and Automation Magazine, vol. 32, no. 2, pp. 35–48, 2025.
- [29]Y. Yang, et al., “Cajun: Continuous adaptive jumping using a learned centroidal controller,” in Conference on Robot Learning. PMLR, 2023, pp. 2791–2806.arxiv.org/abs/2306.09557
- [30]——, “Agile continuous jumping in discontinuous terrains,” in 2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 10245–10252.arxiv.org/abs/2409.10923
- [31]H. Xu, et al., “Dual-horizon hybrid internal model for low-gravity quadrupedal jumping with hardware-in-the-loop validation,” arXiv preprint arXiv:2603.07999, 2026.arxiv.org/abs/2603.07999
- [32]M. Mittal, et al., “Symmetry considerations for learning task symmetric robot policies,” in 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 7433–7439.arxiv.org/abs/2403.04359
- [33]——, “Isaac lab: A gpu-accelerated simulation framework for multi-modal robot learning,” arXiv preprint arXiv:2511.04831, 2025.arxiv.org/abs/2511.04831
- [34]J. Schulman, et al., “Proximal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017.arxiv.org/abs/1707.06347
- [35]N. Rudin, et al., “Learning to walk in minutes using massively parallel deep reinforcement learning,” in Conference on Robot Learning. PMLR, 2022, pp. 91–100.
- [36]F. Bjoljenic, et al., “Towards bridging the gap: Systematic sim-to-real transfer for diverse legged robots,” 2025.
- [37]P. Arm, et al., “Energy-efficient learning-based control of a legged robot in multiple gravity environments,” in 2025 International Conference on Space Robotics (iSpaRo), 2025, pp. 573–579.