Simulation Assessment of Adaptive Communication and Resilient Distributed Model Predictive Control for Robot Swarms under Communication and Sensing Uncertainty

Abstract Physical robot swarms must coordinate under limited communication bandwidth, imperfect sensing, environmental change, and constrained onboard resources. Distributed model predictive control (DMPC) offers a principled way to coordinate constrained agents, but practical deployment depends on how information is exchanged and how local controllers behave when that information becomes delayed or unavailable. We formulate a hierarchical resource-aware DMPC architecture, Swarm-AR-DMPC, with an interpretable event condition based on predicted formation error, safety margin, and information age, together with a stale-information fallback policy. We evaluate the supplied implementation through a controlled simulation study comprising 144 runs: two controller variants, six operating conditions, and 12 random trials per controller-condition pair. The conditions include nominal communication, 10% packet loss with 50 ms delay, 25% packet loss with 100 ms delay, 45% packet loss with 200 ms delay, a dynamic-obstacle case, and sensor degradation. In the present simulation logs, both methods achieved 0/12 task successes in every condition. The adaptive and fixed methods also generated identical communication loads of 440 messages and 21,120 bytes per run, indicating that the current implementation does not yet realize a measurable communication-saving effect. Across all conditions, mean final error was 0.1882 for the adaptive controller and 0.1894 for the fixed controller, while mean formation RMSE was 0.0772 and 0.0769, respectively. These results should therefore be interpreted as a diagnostic simulation assessment rather than evidence of successful physical-swarm deployment. The study identifies the key implementation gaps that must be resolved before a physical experimental validation can support claims of adaptive communication, resilience, and resource efficiency.