Quality Assurance for Autonomous Driving Systems: A Software Engineering Perspective
Tutorial Talk at Autoware Tutorial, 2024 IEEE Intelligent Vehicles Symposium (IV 2024), Jeju Island, Korea
CV
Research Interests
Learning-based Control, Safe Reinforcement Learning, AI/ML in Robotics, AI-enabled Robotic System, Safety and Quality Assurance, Cyber-Physical System
Work Experience
- 07/2022 - Current: Postdoctoral Research Fellow at University of Alberta
- Advisors: Assoc. Prof. Lei Ma, Prof. Zhan Shu
- Developing AI-enabled robotic systems and cyper-physical systems with safety and reliability guarantees
- 08/2017 - 03/2022: Research Associate (Scientific Employee) at Technical University of Munich
- Advisor: Univ. Prof. Dr.-Ing./Univ. Tokio habil. Martin Buss
- Designed general safe reinforcement learning approaches and AI-based control frameworks for robotic systems
- 11/2016 - 06/2017: Research Intern at Technical University of Munich
- Advisor: PD Dr.-Ing. habil. Dirk Wollherr
- Designed learning-based human motion prediction framework for safe human-robot interaction in close-proximity
Education
- 08/2017 - 07/2022: Ph.D. in Electrical and Computer Engineering (Automatic Control and Robotics), Technical University of Munich, Munich, Germany
- 10/2014 - 03/2017: M.Sc. in Electrical and Computer Engineering, Technical University of Munich, Munich, Germany
- 09/2010 - 07/2014: B.Eng. in Mechatronics Engineering, Tongji University, Shanghai, PR China
Research Projects and Collaborations
- 2024 - Current: Quality and Safety Assurance for AI-enabled Autonomous Driving Systems at University of Alberta
- Advisor: Assoc. Prof. Lei Ma
- Developing AI-based perception and control modules for autonomous driving systems
- Integrating quality and safety assurance into AI-enabled autonomous driving systems
- 2024 - Current: AI-enabled Resilient Grid for Clean Energy Integration at University of Alberta
- Advisors: Assoc. Prof. Lei Ma, Prof. Zhan Shu
- Investigating the application of AI approaches in power grids and energy systems
- Developing a testing and evaluation framework for AI-enabled energy systems
- 2023 - Current: Application of Foundation Models in Robotics with Safety Assurance at University of Alberta
- Advisors: Assoc. Prof. Lei Ma, Prof. Zhan Shu
- Developing intelligent robotic systems guided by foundation models, e.g., Large Language Models (LLMs)
- Incorporating safety assurance to robotic systems controlled by cutting-edge AI methods
- 2022 - 2023: Safety and Reliability Assurance of Next Generation AI-enabled Cyber-Physical Systems (AI-CPSs) for Energy Systems at University of Alberta
- Advisors: Assoc. Prof. Lei Ma, Prof. Zhan Shu
- Developed safety analysis approaches for AI-CPSs across various application domains
- Evaluated the performance and reliability of AI components in CPSs
- 2022 - 2023: Benchmarking and Evaluating AI-enabled Cyber-Physical Systems (AI-CPSs) for Robotic Manipulation at University of Alberta
- Advisors: Assoc. Prof. Lei Ma, Prof. Zhan Shu
- Developed a benchmark and an evaluation framework for AI-CPSs in robotic manipulation using NVIDIA Isaac Sim
- Conducted performance and safety analysis for AI-CPSs in diverse robotic manipulation tasks
- 2020 - 2021: AI-based Control Framework for Dual-Arm Robotic Manipulator System at Technical University of Munich
- Advisor: Univ. Prof. Dr.-Ing./Univ. Tokio habil. Martin Buss
- Proposed an AI-based control framework for dual-arm robots, covering the entire perception-planning-control pipeline
- Applied to various real-world dual-arm manipulation tasks
- 2018 - 2020: Safe Reinforcement Learning for Dynamical and Robotic Systems at Technical University of Munich
- Advisor: Univ. Prof. Dr.-Ing./Univ. Tokio habil. Martin Buss
- Proposed general safe reinforcement learning frameworks for complex dynamical and robotic systems
- Applied to diverse robotic systems, including UAVs, robotic manipulators, humanoids, and mobile robots
- 2016 - 2017: Understanding and Predicting Human Reaching Motions for Safe Close-Proximity Human-Robot Interaction at Technical University of Munich
- Advisor: PD Dr.-Ing. habil. Dirk Wollherr
- Designed a learning-based human motion prediction framework for safe human-robot interaction in close-proximity
- Validated and applied to real-world assembly tasks
Professional Activities
Talks
ISR-LLM: Iterative Self-Refined Large Language Model for Long-Horizon Sequential Task Planning
Conference Presentation at 2024 IEEE International Conference on Robotics and Automation (ICRA), Yokohama, Japan
Safe Reinforcement Learning with Model Order Reduction Techniques
Invited Talk at The 1st International Workshop on Safe Reinforcement Learning Theory and its Applications, 2022 IEEE International Conference on Multi-sensor Fusion and Integration (MFI 2022), Cranfield, United Kingdom
Reviewer
- IEEE Transactions on Robotics (T-RO)
- IEEE Transactions on Automatic Control (TAC)
- IEEE Transactions on Neural Network and Learning Systems (TNNLS)
- IEEE Robotics and Automation Letters (RA-L)
- IEEE International Conference on Robotics and Automation (ICRA)
- IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
- IEEE Conference on Decision and Control (CDC)
- European Control Conference (ECC)
- World Congress of the International Federation of Automatic Control (IFAC)
Teaching
Publications
Journal
Enabling Versatility and Dexterity of the Dual-Arm Manipulators: A General Framework Toward Universal Cooperative Manipulation
Yi Ren*, Zhehua Zhou* (equal contribution), Ziwei Xu, Yang Yang, Guangyao Zhai, Marion Leibold, Fenglei Ni, Zhengyou Zhang, Martin Buss, Yu Zheng
IEEE Transactions on Robotics, February, 2024
In this article, we propose a general and versatile control framework for dual-arm manipulators.
PDF
Off-Policy Risk-Sensitive Reinforcement Learning-Based Constrained Robust Optimal Control
Cong Li, Qingchen Liu, Zhehua Zhou, Martin Buss, Fangzhou Liu
IEEE Transactions on Systems, Man, and Cybernetics: Systems, October, 2022
This article proposes an off-policy risk-sensitive reinforcement learning (RL)-based control framework to jointly optimize the task performance and constraint satisfaction in a disturbed environment.
PDF
Learning a Low-Dimensional Representation of a Safe Region for Safe Reinforcement Learning on Dynamical Systems
Zhehua Zhou, Ozgur S Oguz, Marion Leibold, Martin Buss
IEEE Transactions on Neural Networks and Learning Systems, September, 2021
In this article, we propose a general data-driven model order reduction approach for safe reinforcement learning.
PDF
A General Framework to Increase Safety of Learning Algorithms for Dynamical Systems Based on Region of Attraction Estimation
Zhehua Zhou, Ozgur S Oguz, Marion Leibold, Martin Buss
IEEE Transactions on Robotics, June, 2020
In this article, we propose a computationally effective and general safe learning framework, specifically for complex dynamical systems.
PDF
An Inverse Optimal Control Approach to Explain Human Arm Reaching Control Based on Multiple Internal Models
Ozgur S Oguz, Zhehua Zhou, Stefan Glasauer, Dirk Wollherr
Scientific Reports, April, 2018
In this study, we utilize an Inverse Optimal Control (IOC) framework to find the combination of internal models for different human reaching motions.
PDF
A Hybrid Framework for Understanding and Predicting Human Reaching Motions
Ozgur S Oguz, Zhehua Zhou, Dirk Wollherr
Frontiers in Robotics and AI, March, 2018
In this work, we propose a hybrid framework for understanding and predicting human reaching motions for human-robot collaboration.
PDF
Conference
ISR-LLM: Iterative Self-Refined Large Language Model for Long-Horizon Sequential Task Planning
Zhehua Zhou, Jiayang Song, Kunpeng Yao, Zhan Shu, Lei Ma
2024 IEEE International Conference on Robotics and Automation (ICRA), May, 2024
In this work, we introduce ISR-LLM, a novel framework that improves LLM-based planning through an iterative self-refinement process.
PDF
Towards Building AI-CPS with NVIDIA Isaac Sim: An Industrial Benchmark and Case Study for Robotics Manipulation
Zhehua Zhou, Jiayang Song, Xuan Xie, Zhan Shu, Lei Ma, Dikai Liu, Jianxiong Yin, Simon See
IEEE/ACM 46th International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP), April, 2024
As a foundational step towards building reliable AI-enabled robotics systems, in this paper, we propose a public benchmark for robotics manipulation.
PDF
Guided Sequential Manipulation Planning Using a Hierarchical Policy
Hoai My Van, Ozgur S Oguz, Zhehua Zhou, Marc Toussaint
Robotics: Science and Systems (RSS) Conference 2020 - Learning in Task and Motion Planning Workshop, June, 2020
In this work, we introduce a hierarchical policy structure that selects high-level actions for effective task and motion planning (TAMP) in sequential manipulation tasks.
PDF
Hybrid Human Motion Prediction for Action Selection Within Human-Robot Collaboration
Ozgur S Oguz, Volker Gabler, Gerold Huber, Zhehua Zhou, Dirk Wollherr
2016 International Symposium on Experimental Robotics (ISER), March, 2017
In this work, we present a Human-Robot-Collaboration (HRC) framework consisting of a hybrid human motion prediction approach together with a game theoretical action selection.
PDF
Preprint
Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixture
Jiayang Song, Yuheng Huang, Zhehua Zhou, Lei Ma
ArXiv Preprint:2407.07342, July, 2024
In this work, we introduce Multilingual Blending, a mixed-language query-response scheme designed to evaluate the safety alignment of various state-of-the-art LLMs.
PDF
GenSafe: A Generalizable Safety Enhancer for Safe Reinforcement Learning Algorithms Based on Reduced Order Markov Decision Process Model
Zhehua Zhou, Xuan Xie, Jiayang Song, Zhan Shu, Lei Ma
ArXiv Preprint:2406.03912, June, 2024
In this work, we introduce a Genralizable Safety enhancer (GenSafe) to improve the safety performance of safe reinforcement learning algorithms.
PDF
Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forward
Xuan Xie, Jiayang Song, Zhehua Zhou, Yuheng Huang, Da Song, Lei Ma
ArXiv Preprint:2404.08517, April, 2024
In this work, we conduct a comprehensive evaluation of the effectiveness of existing online safety analysis methods on LLMs.
PDF
Self-Refined Large Language Model as Automated Reward Function Designer for Deep Reinforcement Learning in Robotics
Jiayang Song*, Zhehua Zhou* (equal contribution), Jiawei Liu, Chunrong Fang, Zhan Shu, Lei Ma
ArXiv Preprint:2309.06687, September, 2023
In this work, we propose a novel LLM framework with a self-refinement mechanism for automated reward function design.
PDF
Mosaic: Model-based Safety Analysis Framework for AI-enabled Cyber-Physical Systems
Xuan Xie, Jiayang Song, Zhehua Zhou, Fuyuan Zhang, Lei Ma
ArXiv Preprint:2305.03882, May, 2023
In this work, we propose Mosaic, a model-based safety analysis framework for AI-enabled cyber-physical systems (AI-CPSs).
PDF
Data Generation Method for Learning a Low-dimensional Safe Region in Safe Reinforcement Learning
Zhehua Zhou, Ozgur S Oguz, Yi Ren, Marion Leibold, Martin Buss
ArXiv Preprint:2109.05077, September, 2021
In this work, we investigate how different training data will affect the safe reinforcement learning approach based on data-driven model order reduction.
PDF