Results in Control and Optimization (Dec 2024)
A pursuit-evasion game robot controller design based on a neural network with an improved optimization algorithm
Abstract
A pursuit-evasion game (PEG) is a type of game that utilizes one or several cooperative pursuers to capture one or several evaders. The PEG game concept has been used in different multi-robot applications such as transportation or navigation applications, search and rescue, surveillance applications such as collision avoidance and air traffic control systems, multi-defense applications such as missile guidance systems, and medical applications such as analyzing biological behaviors. Regardless of the benefits of PEG, one of the main drawbacks of such systems is the computational burden and the immense time required to learn such systems. For this reason, this work proposes a neural network game based on the pursuit-evasion game, where the leader (evader) robot tries to eat several particles/apples distributed inside a closed game environment with boundary and inner obstacles. In contrast, a follower (pursuer) robot tries to capture the leader robot and stop the particle-eating process. The leader and follower robots were designed based on a differential two-wheel robot (DTWR). The neural network is presented to control and learn the leader and follower robot directions with respect to the boundary and inside obstacles in the game environment. The neural network weights are learned using an improved sine cosine algorithm based on chaotic theory (ISCACT). The ISCACT is proposed to solve and avoid the proposed game of being trapped in the local minimum problem. The ISCACT is tested based on five multimodal benchmark functions. The ISCACT has been used in two cases, the first case arises when ISCACT is used in the follower robot’s learning process. In the second case, the ISCACT has been used in the leader robot’s learning process. The results for the first and second cases prove the superiority of the ISCACT compared with other existing works in enhancing the PEG performance time and reducing the computational burden for multi-robot applications.
