An autonomous system for recovery from object manipulation errors in industrial robot tasks

Abstract

The reliable automatic execution of a robot task depends on the accuracy of the robot manipulator, the correctness of its task description, and the certainty of its environment. The field of robotics has been relatively successful in achieving the first two of these goals. However, real-world uncertainties often interfere with the robot's operation in tasks involving object manipulation. Unexpected events such as misplaced, dropped, or misoriented parts will cause the robot to completely fail at such tasks, no matter how minor the disturbance may be. Recovery from such failures is more difficult than recovery from controller errors because these errors can manifest themselves in many different ways. This thesis presents a robot system architecture that automatically detects part manipulation errors and recovers from them. The system takes a description of the robot's task, converts it into a specially designed internal representation, and provides an execution environment that handles automatic error detection and recovery. The internal representation describes in detail how the robot should operate and how it should use available sensors to monitor the task's execution. Instead of relying on special purpose error recovery code the system exploits knowledge about the robot's task and environment to analyze errors it detects and then to produce appropriate error recovery procedures. A working prototype has been developed that demonstrates the feasibility of this approach. The prototype executes a variety of manipulation tasks on a real robot and automatically handles errors involving lost, dropped, or obstructed parts.