Withdraw
Loading…
Towards failure aware autonomy: robust control, anomaly detection, and robot assistance
Ji, Tianchen
Loading…
Permalink
https://hdl.handle.net/2142/125618
Description
- Title
- Towards failure aware autonomy: robust control, anomaly detection, and robot assistance
- Author(s)
- Ji, Tianchen
- Issue Date
- 2024-07-11
- Director of Research (if dissertation) or Advisor (if thesis)
- Driggs-Campbell, Katie
- Doctoral Committee Chair(s)
- Driggs-Campbell, Katie
- Committee Member(s)
- Chowdhary, Girish
- Mitra, Sayan
- Wang, Shenlong
- 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
- Anomaly Detection
- Machine Learning
- Autonomous Systems
- Abstract
- Robots are entering the open world with an ultimate goal of achieving full autonomy with no human intervention. However, due to complex and uncertain environments, it is hard and nearly impossible to claim that robots will be failure free under any real-world circumstances in the foreseeable future. To make matters worse, such vulnerability grows as robots get equipped with machine learning algorithms, which are brittle in unseen scenarios. Can we make robots reliable and trustworthy even with the presence of possible robot failures? In this dissertation, we try to answer this question by developing failure-aware autonomy that enables robots to prevent, detect, and recover from failures and/or anomalies in real-world applications. As long as failures are acknowledged, identified, and handled appropriately upon occurrence, robots can still deliver robust performance in complex and uncertain environments. To this end, we present approaches for (a) failure prevention via robust control, (b) learning-based anomaly detection in uncertain and interactive environments, and (c) failure recovery through robot assistance from a human supervisor. The key technical developments within these themes include: (1) A novel robust output feedback model predictive controller that guarantees constraint satisfaction and stability under ellipsoidal uncertainty. (2) A state-of-the-art anomaly detector for field robot navigation that can fuse heterogeneous high-dimensional sensor modalities effectively for robust perception and can alert robots proactively before navigation failures occur. (3) A novel unsupervised anomaly detector for autonomous driving that generates a comprehensive anomaly score through an ensemble of neural networks by learning multi-modal normal patterns in driving scenarios. (4) The first algorithm for solving a multi-robot assistance problem as a dynamic graph traversal problem with real-time performance on robot fleets of moderate size. The techniques jointly form failure-aware autonomy and achieve robust performance in real-world applications.
- Graduation Semester
- 2024-08
- Type of Resource
- Thesis
- Handle URL
- https://hdl.handle.net/2142/125618
- Copyright and License Information
- Copyright 2024 Tianchen Ji
Owning Collections
Graduate Dissertations and Theses at Illinois PRIMARY
Graduate Theses and Dissertations at IllinoisManage Files
Loading…
Edit Collection Membership
Loading…
Edit Metadata
Loading…
Edit Properties
Loading…
Embargoes
Loading…