Compliant Manipulation with Contact Awareness and Learning for Safe Physical Human-robot Interaction

Abstract

The growing need for physical human-robot interaction is driven by factors such as an aging population, the desire to automate daily tasks, and the scarcity or prohibitive expense of human expertise in certain domains. However, ensuring safety is paramount to the successful integration of robots into human environments. To attain this objective, extensive research has been dedicated to optimizing robot design, sensing capabilities, and learning algorithms. In light of this background, this thesis presents a new approach to enhancing robot safety by introducing novel compliant actuation designs, an improved learning-based collision detection system, and a collision handling pipeline.

Initially, we present a new compliant actuation system, the discrete variable stiffness actuator (DVSA), which enables rapid stiffness adjustments to pre-defined levels, irrespective of the gaps between stiffness levels. Subsequently, we propose a pipeline that integrates post-collision detection and DVSA-based collision reaction strategy. This pipeline is experimentally evaluated to prove its effectiveness for enhancing safety in the event of unexpected collisions. Furthermore, to enhance the robot’s collision-sensing capabilities, we introduce a learning-based collision detection approach that combines continuous wavelet transform and convolutional neural networks. This method is more sample-efficient, rendering it well-suited to handling insufficient and highly imbalanced datasets - a common issue for learning-based collision detection methods. Additionally, the proposed approach can be applied to random motion, collision location, various intrinsic joint stiffnesses, and various robot joints. With these three parts, this thesis highlights that the integration of compliant actuation, contact sensing, and learning can significantly improve safety during physical human-robot interactions.

Date of Award	Aug 2023
Original language	American English
Supervisor	Irfan Hussain (Supervisor)