Transformer-Based Path Planning for Single-Arm and Dual-Arm Robots in Dynamic Environments

As the manufacturing industry progresses towards more automated and intelligent systems, manipulators are increasingly employed in tasks such as assembly and logistics handling. Traditional path planning algorithms like RRT and RRT* often result in inefficient robot arm operations, generating excessive joint revolutions and large travel distances for the end-effector. Additionally, the sharp turns produced by these algorithms are not suitable for smooth execution of manipulators. This research introduces a novel approach called T-ABA*, which combines the Adaptive Bidirectional A* (ABA*) algorithm with Transformer models to optimize both path planning and obstacle avoidance for robotic arms. The Transformer enhances the path planning process by dynamically adjusting the heuristic functions, improving the identification of efficient routes that avoid obstacles and ensure smooth motion. Furthermore, the Transformer is used to handle dynamic obstacle avoidance by predicting potential collisions and adjusting the path accordingly, reducing computational effort. Additionally, the method is extended to dual-arm control, addressing the complex challenge of collision avoidance between two cooperative robot arms. The proposed approach is validated through computer-based simulations, demonstrating its superiority in optimizing path planning, overcoming obstacles, and preventing inter-arm collisions in both single-arm and dual-arm robotic tasks.