Robust MADER: Decentralized and Asynchronous Multiagent Trajectory Planner Robust to Communication Delay

Abstract

Although communication delays can disrupt multiagent systems, most of the existing multiagent trajectory planners lack a strategy to address this issue. State-of-the-art approaches typically assume perfect communication environments, which is hardly realistic in real-world experiments. This paper presents Robust MADER (RMADER), a decentralized and asynchronous multiagent trajectory planner that can handle communication delays among agents. By broadcasting both the newly optimized trajectory and the committed trajectory, and by performing a delay check step, RMADER is able to guarantee safety even under communication delay. RMADER was validated through extensive simulation and hardware flight experiments and achieved a 100% success rate of collisionfree trajectory generation, outperforming state-of-the-art approaches.

Decentralized Multi-Agent Planning Using Model Predictive Control and Time-Aware Safe Corridors

“…6a) exchanging positions. We compare our method with RMADER [13] and 2 versions of Ego-Swarm [8]. We set the maximum velocity v max = 10 m/s, the maximum acceleration a max = 20 m/s 2 and the maximum jerk j max = 30 m/s 3 for RMADER, Ego-Swarm and our method (along the x, y and z directions).…”

“…For MADER and RMADER, each agent is represented as a bounding box of size 0.25×0.25×0.25 m. For Ego-Swarm and our planner, each agent is represented as a sphere of diameter 0.25 m as per the experiments in [13] (at the time of writing, the bounding box dimensions and sphere diameter were not mentioned in [13], but they were communicated to us by the authors of [13]). The comparison is done with 100 simulated runs for communication latencies equal to 0, 50, and 100 milliseconds.…”

“…Finally, RMADER (Robust MADER) is proposed in [13], which is an extension of MADER [7]. They convexify the optimization problem in order to improve the computation time.…”

“…Our planner and Ego-Swarm [8] use voxel girds as representations of the obstacles in the environment. MADER [7] and RMADER [13] on the other hand use a polyhedral representation of the environment i.e. all obstacles are represented by a series of convex polyhedra.…”

“…This representation is not trivial to generate and may add considerable overhead to the autonomous navigation pipeline. [8], slow Ego-Swarm (Slow ES) [8], MADER [7], RMADER [13] and our method. The comparison consists of 100 simulations with communication delays between 10 agents exchanging positions in a circular configuration as in Fig.…”