LAUSR

Most approaches to multi-robot control either rely on local decentralized control policies that scale well in the number of agents, or on centralized methods that can handle constraints and produce rich system-level behavior, but are typically computationally expensive and scale poorly in the number of agents, relegating them to offline planning. This work presents a scalable approach that uses distributed trajectory optimization to parallelize computation over a group of computationally-limited agents while handling general nonlinear dynamics and non-convex constraints. The approach, including near-real-time onboard trajectory generation, is demonstrated in hardware on a cable-suspended load problem with a team of quadrotors automatically reconfiguring to transport a heavy load through a doorway.