ROTOTILLER: Rolling Optimization of Tasks and Online Trajectories for Integrated Multi-Robot Task Allocation and Fleet Navigation

This paper presents ROTOTILLER (Rolling Optimization of Tasks and Online Trajectories Integrating Local Lookahead Extended Routing), an extension of the extended-SPADES framework for integrated task allocation and fleet navigation in dynamic logistics environments. The approach preserves the coupling between multi-robot task assignment and Covariant Hamiltonian Optimization for Motion Planning CHOMP-based motion planning while introducing rolling-CHOMP, a fleet-level rolling-window optimizer that incrementally updates trajectories in response to new tasks, map changes, and execution disturbances. For task allocation, ROTOTILLER estimates robot–station and station–station travel costs as weighted shortest paths on a Breadth-First Search Medial-Axis Skeleton (BFS-MAS) topological graph extracted from a SLAM-derived occupancy map. In a hospital-like benchmark, this topological cost space reduced average path-computation time from 298.3 ms to 2.08 ms per query, a 143× speedup, while increasing the estimated path length by approximately 21%. In Gazebo simulations with fleets of up to six differential-drive robots, rolling-CHOMP reduced average task-completion time relative to batch multi-CHOMP across the evaluated task-distribution and dynamic-obstacle scenarios by an average of $33.7\%$ for six-robot fleets, while maintaining the measured inter-robot clearance. The system is implemented in ROS 2 and released as an open-source package with reproducible benchmarks.