Sensor-Based Collision Avoidance for Industrial Robotic Arms

Abstract

Robotic automation has evolved to be an irreplaceable part of production lines. Orderly operation of the robots is a key factor. Yet robots often have to deal with unpredictable factors including human beings or possible failures that lead to unexpected behaviours. Collision avoidance mechanisms are therefore of outmost importance to prevent injuries and damage. A major challenge in the creation of a dependable collision avoidance system is the sensory system that could cover relevant parts of the robot and capacitive sensors are a promising solution. The burden is to overcome the nonlinearity and other limitations of the capacitive sensors and harness their potential to this end. It is cumbersome to estimate the proximity of surfaces of the robot from its environment (which could include other robots) from the capacitive readings, so a novel sensing approach is proposed in this thesis. For industrial applications where the motions are well-defined, a pre-recorded capacitive signature can be used to monitor for unexpected changes. In this thesis the capacitive signature of one or more robotic arms will be used to predict and prevent collisions in a robotic workcell. A short training cycle is used to create a signature that is used at runtime to monitor the robot operation. Capacitive electrodes are placed on strategic locations on the robot arms and the surrounding environments and a supervisor computer uses the readings to cease the operation in case of any abnormality. This thesis describes the details of generating the signature from the training data and the runtime software. The supervisor computer provides a pause and/or go signal to the robot(s). The native controller of each arm is kept in place and the only change needed is the ability of each controller to pause the arm at command when a collision is detected and continue from this paused state. This approach requires minimum changes to the existing robotic equipments and programmes. These hardware requirements are widely available on existing controllers. Signature creation is the process of finding the normal pattern of the capacitance readings from all sensors as well as some expected limits allowing for the variations that are to be expected. The algorithms, reasoning, and experimental data are provided throughout the text. The system is tested on a robotic workcell that includes an actively controlled robot and a passive revolute joint. While the algorithm is universal, the suggested hardware has been shown to provide sampling times of down to 20ms, and positional accuracies of ±2mm or better are achieved for the test setup. The thesis also proposes methods to expand the measurement hardware for increased protection and fault tolerance.