Transport Reinforcement Learning based control System (TROLLS)

Drone swarm control in the water-navigation domain is a challenging problem because the drones need to operate in a dynamic environment (changing winds, waves, and currents) and a noisy measurement environment (dust, salt, and mud) leading to non-optimal navigational capabilities. The development of navigation and assembly capabilities is usually accomplished in simulation environments that, at best, approximate operating conditions.  The result is poor generalizability of these solutions to real world situations. Lacking are robust capabilities that enable the deployment of floating bridge systems as a swarm of autonomous drones in a manner that would enhance the viability of missions that involve river crossings. Most drone swarm methods assume a static environment. Drone-swarm solutions for water-based navigation domains are lacking. To address this critical gap in technology Knowledge Based Systems, Inc. (KBSI) proposes to design, develop and demonstrate the Transport Reinforcement Learning based controlSystem (TROLLS) solution, which will enable the Army to deploy water-based drone swarms to autonomously align and assemble river bridges. The proposed TROLLS approach will use multi-level multi-agent (MLMA) reinforcement learning (RL) to learn trajectory planning, guidance, and navigation for self-alignment of multiple robotic bridge components.  The Phase I research will design MLMA and RL methods and demonstrate the technical viability of these methods in addressing the research challenges.