Fault Handling in Robotic Manipulation Tasks for Model Predictive Interaction Control

Abstract

This paper presents a comprehensive framework for robotic manipulation tasks, incorporating systematic fault handling and recovery strategies. The framework leverages model predictive interaction control (MPIC) as a path-following controller to enable dynamic replanning of motion and wrench references. A three-layered architecture divides the control task into decision-making, trajectory planning, and low-level robot control. The Fault Event Pipeline (FEP) is introduced to provide a structured approach for fault detection of pose and wrench errors, diagnosis, and recovery, supporting both forward and backward strategies. The framework integrates fault handling into task planning and execution, offering a unified solution for reliable robotic operations. Experimental validation with a 7-degree-of-freedom Franka-Emika robot demonstrates the framework's ability to handle diverse faults.