Blind collision detection and obstacle characterisation using a compliant robotic arm

Abstract

This paper presents a novel blind collision detection and material characterisation scheme for a compliant robotic arm. By the incorporation of a simple MEMS accelerometer at each joint, the robot is able to detect collision, identify the material of an obstacle, and create a map of the environment. Detailed hardware design is provided, illustrating its value for building a compact and economical robot platform. The proposed method does not require the additional use of vision sensor for mapping the environment, and hence is termed as `blind' collision detection and environment mapping. Based on the shock wave and vibration signals, the proposed algorithm is able to classify a range of materials encountered. Detailed laboratory evaluation was performed with controlled obstacle collision from different orientation and locations with varying force and materials. The proposed method has achieved 98% detection sensitivity while maintaining 77% specificity. Furthermore, by using sound feature extraction and machine learning techniques, the classifier produces an accuracy of 98% for classifying four different impact materials. In this paper, we also demonstrate its use for detailed environment mapping by using the proposed method.