DDPG-Based Pursuit-Evasion Games of Scale-Variable Multi-Robot Systems Using Zero-Padding

In this article, we focus on pursuit-evasion games of scale-variable multi-robot systems using Deep Deterministic Policy Gradient (DDPG) algorithm. There are some pursuers incoming or exiting pursuit-evasion games, leading to scale-variable issue. To address the variable scale, zero-padding is used in construct critic and actor neural networks (NNs), which can deal with variable input dimensions. A new reward is defined according to requirements of pursuit-evasion games, which can train critic and actor NNs to make the total reward gradually converges in DDPG algorithm. To verify the performance of the proposed method, numerical results are provided and analyzed.