Safe shared autonomy and evaluation metrics

In this work, we want to improve teleoperation safety for a double-wheeled humanoid robot using shared autonomy. We divide the development of this shared autonomous system into three stages: modeling, offline evaluation, and online planning and control. The modeling stage consists of approximating the dynamics of the actual robot hardware. We provide three layers for said approximation: the cartpole, the differential drive, and the double-wheeled inverted pendulum (DWIP) models. Each approximation serves a different purpose in offline evaluation and online planning and control strategies. The offline evaluation stage is the most novel part of this thesis. We present an approach to quantitatively measure the ability of a shared control system to track user input while simultaneously ensuring safety. This is achieved through what we call CPI metrics, based on reachability theory, which measure the Conservativeness (C), Permissiveness (P) and the amount of Intervention (I) applied to a user nominal control. Said metrics apply to arbitrary dynamic systems and state and control constraints, and apply to non-differentiable shared controllers including controllers implemented in procedural code, unlike other existing shared autonomy metrics. Moreover, we propose a parallel algorithm based on Rapidly-exploring Random Trees (RRTs) for conducting the reachability analysis necessary for computing conservativeness and permissiveness metrics efficiently. We consider a Linear Quadratic Regulator (LQR) controller and two different Model Predictive Controller (MPC) based controllers which vary in their approach to approximate the idealized and in practice unsolvable goal of achieving minimum intervention shared control (MISC): MPC-SF obtains the direct control from the user or the control from an unsafe controller such as an LQR and adopts a 1-step loss, while the MPC-TT formulation aggregates the penalization of the target deviation over a given horizon. Evaluating the above controllers on the aforementioned CPI metrics then helps quantify the trade-offs of the respective control strategies between performance and safety, for our specific double-wheeled humanoid robot problem. Finally, in order to make the shared autonomous system able to run online in complex environments, we propose an obstacle-aware strategy, MPC-OA, consisting of an optimization-based trajectory planning layer that is an extension of MPC-TT taking into account environmental constraints, followed by a low-level higher-fidelity LQR tracking layer.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Zhou, Yu

Contributors dc:contributor

• Hauser, Kris

• Self-Balancing Robot
• Robot Safety
• Performance Evaluation Metrics
• Telerobotics and Teleoperation
• Trajectory Optimization
• Collision Avoidance

Statement dc:rights

• Copyright 2022 Yu Zhou

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/115622

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F115622