Cooperative Control of Mobile Robots with Stackelberg Learning

Multi-robot cooperation requires agents to make decisions that are consistent with the shared goal without disregarding action-specific preferences that might arise from asymmetry in capabilities and individual objectives. To accomplish this goal, we propose a method named SLiCC: Stackelberg Learning in Cooperative Control. SLiCC models the problem as a partially observable stochastic game composed of Stackelberg bimatrix games, and uses deep reinforcement learning to obtain the payoff matrices associated with these games. Appropriate cooperative actions are then selected with the derived Stackelberg equilibria. Using a bi-robot cooperative object transportation problem, we validate the performance of SLiCC against centralized multi-agent Q-learning and demonstrate that SLiCC achieves better combined utility.

I. Introduction

Robotics at large has been improving at a rapid pace, and this has resulted in increased demand in applications ranging from manufacturing [1], to warehousing [2], to human-populated environments [3]. However, despite the clear potential for distributed controllers that leverage cooperation between multiple mobile robots (e.g., the cooperative transport of large or heavy objects, see Fig. 1b), the vast majority of existing techniques are either limited to single-robot operation or require that each robot perceive the complete state of the environment [4]. Interestingly enough, model-free learning-based methods present a promising alternative to traditional model-based control, in that they are less reliant on domain knowledge such as kinematic and dynamic modeling of the system, and that they scale more naturally with the number of agents.