Planning of the Coordination of Multiple Quadrotors Applied to the Transport of Materials

Abstract

The problem of resource allocation is still a large study area, where different techniques are applied to find an optimal or sub-optimal solution to the problem. This work presents a solution to this problem that uses a Reinforcement Learning technique called Learning Automata, in conjunction with the A* heuristic search algorithm, to allocate material transport tasks to multiple agents and calculate routes to perform these tasks. The vehicles used as agents are small quadrotors. The A* algorithm was applied to generate optimal local routes for each carrier and occasionally resolve conflicts between them. Diagonal distance heuristics were used and a small modification was made to the algorithm that avoids convergence to a non-optimal route. A Pure Pursuit path tracking algorithm was used to give velocity commands to the agents in order to follow the path reference given by the A* algorithm. The various analyzed cases of the learning algorithm and a scalability test showed that the proposed solution is capable of finding sub-optimal solutions in a reasonable time for small and medium scale problems, showing that the route plan learned can solve the proposed tasks. The solutions were applied in the Gazebo simulation environment where the communication with the learning algorithm on MATLAB has been done via ROS.