Abstract
Background: Decentralized task allocation in robotic swarms must contend with heterogeneous agent capabilities, exogenous disturbances, and dynamic communication topologies. Classical market-based and consensus-based methods offer scalability but are sensitive to model uncertainty and time-varying conditions. Objective: We develop DARTS, a Decentralized Adaptive Robust Task Swarm allocation framework that couples a provably stable adaptive low-level controller with a consensus-regularized auction mechanism to achieve robust allocation in heterogeneous multi-robot teams. Methods: Each robot maintains local adaptive estimates of its task-specific productivity and execution cost using a Lyapunov-based update with projection and sigma-modification. These estimates parameterize marginal utilities in a consensus-based bidding scheme that converges to a conflict-free assignment under intermittent connectivity. A robustifying sliding term compensates bounded disturbances in the execution layer, while control allocation along the actuator null space mitigates incipient faults. We analyze input-to-state stability of the closed-loop and monotonic improvement in allocation welfare under mild connectivity and dwell-time conditions. Results: In simulations with 60 heterogeneous robots (ground and aerial) and 120 stochastic tasks, DARTS achieved 92.4% mean completion, 3.1% regret, and 12% lower energy per task than consensus auctions and 18–22% lower than market-based and evolutionary baselines. Parameter estimates converged to neighborhoods whose radii scaled with disturbance bounds; allocation fairness improved relative to baselines. Ablations confirmed that both adaptive estimation and robust control are necessary for performance under high heterogeneity and switching graphs. Conclusions: By unifying decentralized adaptive control with allocation, DARTS provides strong performance and robustness without centralized coordination. The approach complements established market/auction paradigms and extends adaptive control ideas to the multi-robot allocation layer, offering a principled path toward resilient heterogeneous swarms in dynamic environments.
Keywords
decentralized adaptive control, task allocation, heterogeneous multi-robot systems, robustness, consensus-based auctions, sliding mode compensation, fault-tolerant allocation, swarm robotics