Crucial Factors Affecting Decentralized Multirobot Learning in an Object Manipulation Task

Decentralized multirobot learning refers to the use of multiple learning entities to achieve the optimal solution for the overall robot system. We demonstrate that single-robot learning theory can be successfully used with multirobot systems, but with certain conditions. The success and the effectiveness of this method are potentially affected by various factors that we classify into two groups: the nature of the robots and the nature of the learning entities. Incorrect setup of these factors may lead to undesirable results. In this paper, we methodically test the effect of varying four common factors (reward scope, learning algorithms, diversity of robots, and number of robots) in a decentralized multirobot system, first in simulation and then on real robots. The results show that two of these factors, reward scope and learning algorithm, if set up incorrectly, can prevent optimal, cooperative solutions.