Learning structured interaction models for robot navigation in human environments

Liu, Shuijing

Permalink

https://hdl.handle.net/2142/125626 Copy

Description

Title

Learning structured interaction models for robot navigation in human environments

Author(s)

Liu, Shuijing

Issue Date

2024-07-12

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

Driggs-Campbell, Katherine

Doctoral Committee Chair(s)

Driggs-Campbell, Katherine

Committee Member(s)

• Amato, Nancy M.
• Hauser, Kris
• Gupta, Saurabh

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Robotics
• Machine Learning
• Human-robot Interaction
• Robot Navigation
• Artificial Intelligence.

Language

eng

Abstract

Robots are becoming increasingly prevalent in our daily lives. However, these autonomous agents work best in isolation. Integrating robots into human environments remains challenging due to the complex and dynamic nature of human-robot interactions. This thesis addresses the challenge of robot navigation in human environments by proposing structured learning methods that enhance robots’ ability to interact with humans in a safe, efficient, and socially aware manner. We explore two types of interactions: implicit interactions through motion and explicit interactions through language. We build navigation systems that predict human intentions, reason about subtle interactions among agents, and plan paths for robots to fulfill tasks. In previous work, model-based approaches and end-to-end learning methods have demonstrated their own advantages and limitations. To combine the best of both worlds, this thesis develops structured ML systems by incorporating the predictions of human behaviors into robot planning. In addition, we formulate interactive scenarios using graph structures, which enables robots to achieve a more nuanced understanding of human behavior and interaction dynamics. We propose intention and interaction-aware robot decision-making systems for navigation in human spaces. Through simulation benchmarks, real world experiments, and user studies, we demonstrate robot’s capability to navigate in various complex human environments. Our contributions span three applications: (1) driver trait inference for autonomous vehicle navigation at T-intersections, (2) crowd navigation using attention-based spatio-temporal graphs, and (3) assistive navigation for persons with visual impairments using dialogue and visual-language grounding. Through these applications, we demonstrate the effectiveness of structured learning in robotic tasks that involve human-robot interaction. This thesis proposes algorithms and tools for deploying robots in real-world human settings, advancing the field of learning-based human-robot interaction.

Graduation Semester

2024-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/125626

Copyright and License Information

Copyright 2024 Shuijing Liu

Owning Collections