betaFeedback

WorksEngineering

Temporal Attention-Enhanced Mixture-of-Experts Guidance for Multi-Revolution Low-Thrust Rendezvous

Abstract

Multi-revolution low-thrust rendezvous requires both long-duration orbital transfer and precise terminal control. This paper proposes a temporal attention-enhanced mixture-of-experts guidance law trained by deep reinforcement learning. A far-range expert acts on modified equinoctial elements, a terminal expert acts on the relative state in the target's local frame, and a learned gate blends their thrust commands. The terminal expert encodes its observation history with a Transformer to address partial observability, and the gate uses a separate encoder to account for the resulting history-dependent command. The experts are pretrained separately, the gate is trained with both experts frozen, and the complete guidance law is then fine-tuned end to end. On 600 test missions, it achieves rendezvous in every case at tolerances of 10 m and 1 m/s, reduces mean flight time and delta-v by 50.6% and 44.5% relative to distance-based hard switching, and remains successful without retraining under navigation errors, unmodeled orbital perturbations, eclipse-induced thrust outages, and thrust loss.

Paper details

vlnc:W_vxvyt8xc
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 0

downloads · 0

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

  • Engineering

AI-assisted tools

View statement
Looking for related papers…