Learning from physical human-robot interaction with velocity-controlled instantaneous responses and update thresholds for noise rejection

Xie, Yiqing

Permalink

https://hdl.handle.net/2142/120169 Copy

Description

Title

Learning from physical human-robot interaction with velocity-controlled instantaneous responses and update thresholds for noise rejection

Author(s)

Xie, Yiqing

Issue Date

2023-05-03

Director of Research (if dissertation) or Advisor (if thesis)

Driggs-Campbell, Katherine Rose

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• pHRI
• Robotics
• Intention Recognition

Abstract

The physical human-robot interaction (pHRI) is an important means for humans to control robots in real-time. However, human touch may or may not be intentional. Current approaches react to these force inputs regardless of whether they are intended to be meaningful. The robot will also retain its behavior logic after the human’s interference. Recent research has shown that robots can learn from pHRI and adjust their behavior logic in real-time. We replicate and extend a state-of-the-art impedance-controlled approach to the pHRI problem on a platform without torque control. We believe that the data generated by pHRI can reveal a user’s true in-tentions, such as waypoints to pass or obstacles to avoid. To examine the connection between pHRI and intent, we first create an experimental testbed for pHRI with a UR5e platform. Then, using the data we collect, we estimate the human’s intentions from the force input and translate these intentions into potential features for the robot to learn. Through these steps, the robot can understand the human’s goal. Extending prior art in this type of learn-ing, we propose a new update rule for the robot to learn human intentions that takes into account unintentional forces, making the learning process more robust.

Graduation Semester

2023-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2023 Yiqing Xie

Owning Collections