Explainable AI for Obstacle-Aware Robotic Manipulation

This paper addresses the critical need for explainability in AI-driven robotics, particularly in inverse kinematics (IK). It proposes a methodology to make neural network-based IK models more transparent and safer by integrating Shapley value attribution and physics-based obstacle avoidance evaluation. The study focuses on the ROBOTIS OpenManipulator-X and compares different IKNet variants, providing insights into how architectural choices impact both performance and safety. The work is significant because it moves beyond just improving accuracy and speed of IK and focuses on building trust and reliability, which is crucial for real-world robotic applications.