A fault-tolerant collision avoidance architecture for autonomous excavators using digital twin-based predictive and real-time detection

We propose a collision avoidance architecture for excavators that satisfies fault tolerance by in-machine and on-site detection. Safety on construction sites is essential to avoid fatal accidents or damage to machinery. A fault-tolerant design is beneficial to safety because it allows necessary countermeasures to be taken in case of partial problems. However, one of the challenges is the consideration of delays caused by hydraulic characteristics, mechanisms, communications, and processing in outdoor construction machinery. Another difficulty is that collision detection with a safety margin to account for various delays may reduce work efficiency due to excessive detection. Therefore, we propose dual Digital Twin-based Collision Check (DTCC) systems to ensure safety without reducing work efficiency. The system ensures safety with two components. One is a detection function processed within the machine based on the current state to improve the certainty of stopping. The other is the ability to predict machine motion based on control inputs and intervene to prevent hazards. Through numerical experiments and verification using a real excavator, we confirmed that the combination of early deceleration by predictive detection and emergency stop can reliably stop the excavator without increasing the safety margin, demonstrating the practicality of the proposed method.