Segregation of Wheeled Mobile Robots Using Benchmark Reinforcement Learning Algorithms

The coordination and assignment of individual missions to multiple independent robots is the central focus in the field of robotics. This paper explores how a team of five wheeled mobile robots (WMRs) can be segregated into two separate groups, with each group assigned a specific goal location to reach. The problem is framed within the context of reinforcement learning (RL). Four standard deep reinforcement learning (DRL) algorithms designed for continuous control problems, deep deterministic policy gradient (DDPG), twin-delayed deep deterministic policy gradient (TD3), soft actor–critic (SAC), and proximal policy optimization (PPO), are adopted and simulated. The robots are trained within a simulation environment, and performance is evaluated based on the success rate of the segregation, convergence time, average reward, and the characteristics of the learned control actions (i.e. linear and angular velocities). The results provide a quantitative basis for selecting suitable control algorithms in a multi-robot sorting and task allocation scenario.