Collision Detection for Robot Arm Assisted with Digital Twin System

In industrial and disaster recovery operations, robot arms take over tasks in hazardous areas. Collision detection methods are essential to prevent damage to the surrounding environment and the robot arm itself. Traditional approaches, such as environmental cameras and tactile sensors, face limitations like installation challenges and high costs. Model-based and data-driven strategies also struggle with accuracy, hindered by external disturbances like friction and the need for extensive data collection. Addressing these challenges, this paper introduces a digital twin-based approach for more accurate collision detection. Digital twins, virtual replicas of physical environments, enable cost-effective and safe data collection. This research utilizes a physical simulator within the digital twin to simulate the efforts of a real robot arm based on the robot’s physical characteristics and motion trajectory, enhancing detection accuracy by adjusting for modeling errors and disturbances through linear regression. Experiments with the xArm 7 robot arm, simulating collisions by applying manual force, demonstrate the proposed method outperformed conventional model-based and machine-learning methods.