Memory-Efficient Meta-Reinforcement Learning for Adaptive Safety-Critical Control in Adversarial Spacecraft Proximity Operations

In three linesComparative study of three recurrent architectures (LSTM, GRU, Mamba) and two algorithms (PPO, SAC) for meta-reinforcement learning applied to input-constrained control barrier functions (ICCBF) in spacecraft proximity operations. Mamba + PPO outperforms other setups in safety, task completion, and fuel savings across cooperative and adversarial scenarios.

Summary generated by Claude — human-verified