Skip to Main Content
This paper investigates the problem of object transportation, particularly pushing or moving an object to a goal location and orientation, using multiple robots. A multi-agent architecture is established to realize effective cooperation between multiple autonomous intelligent robots, in carrying out the task. Machine learning is incorporated into the architecture. In the developed approach, the world state of the task is established by fusing sensory information. Two machine learning and optimization methods, reinforcement learning (RL) and genetic algorithms (GA), are combined to learn a cooperation strategy and based on which, determine the optimal actions to reach the task goal. The outputs of RL and GA are evaluated by an arbitrator using a probabilistic method, which resolve conflicts and improve the overall performance. The feasibility of the scheme is illustrated through computer simulation.