Modeling and Sensing of Continuum and Parallel Robots for Physical Human-Robot Interaction
Orekhov, Andrew Leonidovich
0000-0003-3803-2326
:
2022-01-10
Abstract
Increasingly, robots are being designed with compliance to provide robustness to unexpected contact forces, improved force control in the presence of geometric uncertainties, and improved safety in human-robot interaction. Compliance, however, comes at a cost of increased kinematic uncertainty, sensing difficulties, and modeling complexity. To help address these limitations, this dissertation makes contributions in the areas of design, modeling, shape sensing, and control of compliant parallel robots and continuum robots in the context of collaborative manufacturing, with a focus on collaborative manufacturing in confined spaces. We first present a redundancy resolution method for online stiffness modulation of parallel robots utilizing both kinematic redundancy and variable stiffness actuators while avoiding kinematic singularities, joint limits, and leg collisions. We then present the design of a continuum robot module for a new collaborative robot that is a first step towards achieving reconfigurable continuum robots with whole-body sensing. The segment is designed with high torsional stiffness and actuation components that are integrated into the base of the segment. We also present a Lie group kinematic modeling approach using a modal shape basis on the backbone curvature that allows for shape sensing with general string encoder routing. The kinematic model allows for flexibility in the order of the modal basis, bridging the gap between constant-curvature models and more general deflection models. We discuss numerical methods for solving the shape sensing problem and propose methods to optimize the string routing paths to improve numerical conditioning. We then validate the shape sensing approach experimentally on a collaborative continuum segment and in a simulation study for a robot with torsional deflections and helical string routing. Building on this kinematic formulation, we then apply orthogonal collocation on Lie groups to computing the statics of Cosserat rod models and show how the Lie group formulation leads to an analytic expression for the task-space and configuration-space compliance matrices of variable curvature tendon-actuated continuum robots. These methods set the stage for more accurate control of these robots and provide the necessary compliance formulation for future improvements in active compliance control for physical human-robot interaction.