DC-Ada: Reward-Only Decentralized Sensor Adaptation for Heterogeneous Multi-Robot Teams
Saad Alqithami · IEEE Access · 2026
Heterogeneity is a defining feature of deployed multi-robot teams: platforms often differ in sensing modalities, ranges, fields of view, and degradation/failure patterns. A persistent deployment challenge is that controllers trained under a nominal sensing configuration (often homogeneous) can degrade sharply when executed on robots with missing or mismatched sensing, even when the task and action interface remain unchanged. This paper presents DC-Ada, a reward-only decentralized adaptation method that keeps a pretrainedshared policyfrozen and instead adapts compact, per-robotobservation transformsto reconcile heterogeneous sensing into a fixed inference interface. DC-Ada is gradient-free and communication-minimal: adaptation proceeds via budgeted accept/reject random search, using short common-random-number rollouts to compare candidate perturbations under a strict environment-step budget. We evaluate DC-Ada and four baselines (shared policy, observation normalization, random perturbation, and local fine-tuning) in three multi-robot domains—warehouse logistics, search and rescue, and collaborative mapping—across four heterogeneity regimes (H0–H3) and five random seeds under a matched interaction budget. We report shaped return, thresholded task completion, continuous progress metrics, and threshold-sensitivity analyses, together with runtime and communication accounting. Across domains, heterogeneity substantially impacts a frozen shared policy, and the most effective mitigation strategy is task-dependent: simple normalization can improve reward robustness in some settings, while gradient-based fine-tuning can be strong when gradients are stable and compute is available. DC-Ada provides a complementary operating point: it improves completion performance most clearly in coverage-based mapping under severe heterogeneity (including H3), while requiring only scalar team returns and avoiding policy fine-tuning or persistent message exchange. These results position DC-Ada as a practical deploy-time interface adaptation mechanism for heterogeneous teams when gradients, privileged state, or high-bandwidth communication are unavailable or undesirable.