AUTONOMOUS MISSION-PLANNING AND COLLABORATION BETWEEN MULTI-DOMAIN ROBOTS

Abstract

Autonomous multi-robot path-planning and task-allocation to address unresponsive or even evasive targets on or under the water is studied. Challenges in robot localization and navigation in GPS-denied and communication constrained environments are addressed. The mRobot node developed autonomously path-plans and task-allocates to manage robotic collaboration in marine environments. The node is validated in simulation and controlled in-water tests with a team of heterogeneous marine robots (unmanned surface vehicle (USV), unmanned aerial vehicle (UAV), unmanned underwater vehicle (UUV)) to survey static floating targets (of known pose) in three-dimensions. Then, a novel mutual-information incentivised Q-Learning algorithm is developed for UUVs to search for static underwater targets of uncertain pose. The planning considers the complex underwater environments where erroneous and false detections are expected. Simulation and controlled two-dimensional experiments show the algorithm performs notably better than alternative methods like greedy or Boustrophedon. Lastly, a collaborative team of three UUVs is proposed to acoustically detect, track, and localize a mobile, evasive underwater target with uncertain pose. A novel algorithm combines predictive information measures with Q-Learning for trajectory planning. The algorithm adapts to conditions that impact detection with acoustic range-only measurements. Simulation results show superior performance of the 3- UUV system compared to the long baseline method.