Search papers, labs, and topics across Lattice.
SafeDiSTL decomposes joint constraint violations in cooperative MARL by leveraging Signal Temporal Logic (STL) robustness to measure episode-level costs and compute action sensitivities. These sensitivities, derived from differentiable dynamics and smooth full-episode robustness, enable credit assignment across agents for cost increments, even accounting for delayed effects. The framework significantly improves safe-success rates and throughput in cooperative payload transport, connected navigation, and networked traffic control tasks compared to existing constrained baselines.
Joint safety violations are now traceable to individual agents, unlocking robust credit assignment in cooperative MARL even with team-level cost functions.
Cooperative multi-agent tasks often impose constraints on a shared objective or collective safety. A team level violation cost function can measure this constraint outcome, but does not distinguish which agents'actions most strongly influence that violation. We present SafeDiSTL, a constrained multi-agent reinforcement learning framework that allows for credit assignment even with joint constraint measurements. We use Signal Temporal Logic (STL) robustness to measure episode-level violation through running-maximum cost increments. Smooth full-episode robustness and differentiable dynamics provide action sensitivities, which allocate each cost increment across agents at the same decision time. These sensitivities account for delayed effects when weighting agents'shares of each cost increment. We study the framework in cooperative payload transport, connected navigation, and networked traffic control tasks. SafeDiSTL achieves 1.6-5.1x the safe-success rate in transport, improves navigation safe-success by 11.1%, and delivers 50-68% more throughput in traffic when compared to constrained baselines. Results demonstrate a strong improvement of the learned policy in dealing with shared constraint while retaining the performance of unconstrained policies.
