Trustworthy safety-critical systems
Risk perception, cognition, decision-making, and runtime safety.
Heye Huang
Assistant Professor
Cho Chun Shik Graduate School of Mobility, KAIST
I lead ACES Lab (Autonomous Cognitive and Embodied Systems Lab) at KAIST.
Before joining KAIST, I was a Postdoc Associate at the MIT SMART Center, working with Prof. Jinhua Zhao in the MIT JTL Urban Mobility Lab. I was also a Research Associate at the University of Wisconsin-Madison, working with Prof. Xiaopeng (Shaw) Li in the CATS Lab.
I received my Ph.D. in Mechanical Engineering from Tsinghua University, where I was supervised by Prof. Jianqiang Wang and Prof. Keqiang Li in the THICV Lab.
I was also a visiting scholar in Cognitive Robotics at TU Delft, working with Prof. Arkady Zgonnikov and Prof. David Abbink, and gained research experience through internships at UC Berkeley and TUM.
Email / Scholar / ResearchGate
Research
Building reliable autonomy for real-world mobility and robotics.
I am broadly interested in building safe, adaptive, and trustworthy autonomous systems (e.g. AVs, Robotics) that can reason under uncertainty and operate in complex real-world environments. My research currently focuses on three closely connected directions:
Generative AI and world models
LLMs, diffusion models, and multimodal models for prediction, reasoning, and long-tail scenario generation.
Foundation models and agents
Memory-augmented and model-driven agents for embodied intelligence.
Across these directions, my research connects fundamental questions in safety, prediction, and decision-making with deployable methods for real-world autonomy. By integrating risk-aware reasoning, generative models, world models, and foundation-model agents, I seek to build autonomous systems that are adaptive, trustworthy, and effective across driving, robotics, and broader interactive mobility settings.
News
Oct/2026
Benwang Chen’s “AnyHand-Agent” received First Place in the poster presentation at the Nature Conference: Autonomous Robotics.
Aug/2026
- “CAPE-3D: Context-Adaptive Planned-to-Executed 3D World Models for Safer Robot Interaction” received the Best Presentation Award at the IEEE CASE 2026 Workshop on Safety-Critical Robotics.
- “XMind: Risk-Aware Episodic Memory for Continually Improving Safety in Autonomous Driving” received the Best Poster Award at the IEEE CASE 2026 Workshop on Safety-Critical Robotics.
Jun/2026
- “SMART: Scalable Multi-Agent Reasoning and Trajectory Planning in Dense Environments” accepted by Transportation Research Part C: Emerging Technologies.
- “Risk-Controllable Multi-View Diffusion for Driving Scenario Generation” Best Paper Award · Oral Presentation, CVPR 2026 Workshop on VGBE.
- “RESPOND: Risk-Enhanced Structured Pattern for LLM-driven Online Node-level Decision-making” accepted by Communications in Transportation Research.
- “SENSE: Satellite-based ENergy Synthesis for Sustainable Environment” KDD 2026 AI4Sciences Track Oral.
Apr/2026
Joined KAIST as a tenure-track assistant professor.
Selected Awards & Recognitions
2026
- Award for Outstanding Contribution 15th Asia Pacific Transportation and the Environment Conference
2024
- Beijing Municipal Science and Technology Progress Award "First Prize" Beijing Municipal Government
2023
- Best Paper Award IEEE DSInS
- Distinguished PhD Graduate Awardee Ministry of Education of Beijing
- "Academic Rising Star" Award School of Vehicle and Mobility, Tsinghua Univ.
- Best PhD Thesis Award finalist National PhD Academic Forum of the China Society of Automotive Engineers
2022
- ASCE Best Research Award for Risk and Uncertainty in Engineering Systems
- Infineon Innovation Scholarship Infineon Technology
- Journal Cover Paper Award Journal: Engineering
- Comprehensive First-Class Scholarship Tsinghua University
- Doctoral Forum Outstanding Thesis Award Tsinghua University
2021
- Best Paper Award Journal: Accident Analysis & Prevention
- Comprehensive First-Class Scholarship Tsinghua University
- Doctoral Forum Outstanding Thesis Award Tsinghua University
2020
- National Scholarship for Ph.D. Students Ministry of Education of China
2019
- Youth Leadership Development Programme ITS World Congress, Singapore
- Comprehensive First-Class Scholarship Tsinghua University
- Doctoral Forum Outstanding Thesis Award Tsinghua University
Join ACES Lab · KAIST
Open Positions
We are recruiting students and welcoming collaborators interested in foundation models for autonomous driving, world models, embodied AI, and memory-augmented and self-evolving LLM agents.
Research questions
Start with a problem, not just a model
These directions are broad themes rather than fixed project assignments. Specific projects grow from concrete research questions and each student’s strengths.
01
Foundation models for autonomous driving
How can foundation models enable robust perception, reasoning, and decision-making for autonomous driving in complex, open-world, and safety-critical environments?
02
World models as data engine
How can world models generate diverse, realistic, and controllable scenarios for simulation, evaluation, and training in autonomous systems?
03
Embodied AI for autonomous and interactive systems
How can embodied AI systems perceive, reason, and act effectively in dynamic real-world environments, especially for long-horizon tasks, interaction, and adaptation?
04
Memory-augmented and self-evolving LLM agents
How can memory, continual learning, and self-improvement mechanisms help LLM agents adapt over time, accumulate knowledge, and become more reliable in long-horizon tasks?
A candid note
Your first message does not need to be a finished proposal
Across driving and robotics, the recurring question for us is not simply whether a model can produce an action, but whether the system can recognize when its assumptions are failing and adjust safely.
Start with one problem you genuinely care about. Show us how you have approached a difficult question before, what you contributed, and what you would test next. A precise question is more useful than a long list of AI keywords.
If one of our recent papers makes you curious—or if you think an assumption or failure case is missing—tell us why. It is also fine to be explicit about what you do not know yet and want to learn.
Opportunities
Ways to work with us
PhD / Master students
Fall 2027 admission
Tell us which research direction interests you, the question you hope to investigate, and the experience that prepares you for graduate research. CSC-supported PhD applicants are especially welcome.
Research interns and RAs
Remote or on-site
Include your proposed dates, weekly availability, relevant technical skills, and a focused project or research question that overlaps with the lab.
Visiting students and research collaborators
Project-based collaboration
Describe the shared problem, the perspective or resources each side can contribute, and a realistic first step for exploring the collaboration.
A lab inquiry does not replace the official KAIST admissions process. Fall 2027 PhD / Master applicants should also follow the applicable KAIST application requirements and deadlines.
Email application
Contact ACES Lab
Send a short email with your CV, research interests, and intended start date or availability.
Suggested subject
[PhD / Master Fall 2027] Name — Current Institution
Include
- Your CV
- A brief note on your research interests
- Your intended start date or availability
For internships, RA positions, visits, or collaborations, use
[Research Inquiry] Name — Position — Institution and
include your proposed dates and availability.
Before you apply
Brief FAQ
Do I need an exact background match?
No. Relevant strengths may come from machine learning, robotics, control, optimization, simulation, data systems, or related fields. What matters is a clear research connection and evidence that you can contribute to it.
Does sending an email guarantee a position?
No. Availability depends on research fit, timing, and the type of opportunity. Degree admission decisions also follow KAIST’s official process.
What makes an inquiry useful?
A specific research question, a concise explanation of fit, and direct links to representative work are more helpful than a general message sent to many labs.
ACES Lab
Autonomous Cognitive and Embodied Systems Lab
Current Students
PhD Students
- Education M.S., Electronic Information — Tsinghua University
- Research Interests Computer Vision · Deep Learning · Multimodal Learning
- Education M.Sc., Computer Science — EPFL B.Eng. (Hons), EIE — PolyU
- Research Interests Embodied AI · In-Context Learning · Vision-Language Learning
- Education M.Sc., AI — University of Macau B.Eng., Software Engineering — HIT
- Research Interests Safety-Critical Embodied Systems · LLM/VLM Agents · Autonomous Driving
- Education M.S., Cyberspace Security — Beihang B.S., Information Security — Beihang
- Research Interests World Models · Autonomous Driving · Embodied AI
- Education Bachelor’s, Automation — Zhejiang University
- Research Interests Trustworthy Autonomous Driving · Multi-Agent Systems
Visiting Students, Research Interns & RAs
Yuhang Wang
Since Dec. 2024
Weining Ren
Since Feb. 2026
Lingxuan Zhou
Since Apr. 2026
Jienan Lv
Since Apr. 2026
Shaojie Gu
Since Sep. 2026
Yufei Long
Since Sep. 2026
Sean Hardesty Lewis
Jun. 2026 - Oct. 2026
Haiming Tang
Jun. 2026 - Oct. 2026
Jaein Reuw
Jul. 2026 - Sep. 2026
Jingtao Shen
May 2026 - Aug. 2026
Zhiyuan Zhou
Jun. 2024 - Dec. 2025
Yiwei Shu
Aug. 2024 - Aug. 2025
Zhaohui Wang
Nov. 2024 - Aug. 2025
Yibin Yang
Sep. 2023 - Apr. 2024
Publications
(* corresponding author, # equal contributions)
A knowledge-augmented dataset of high-risk driving scenarios with LLM annotations for autonomous driving
SENSE: Satellite-based ENergy Synthesis for Sustainable Environment
Oral Presentation
Risk-Controllable Multi-View Diffusion for Driving Scenario Generation
Best Paper Award · Oral Presentation
RESPOND: Risk-Enhanced Structured Pattern for LLM-driven Online Node-level Decision-making
CogDrive: Cognition-Driven Multimodal Prediction-Planning Fusion for Safe Autonomy
Learning from Risk: LLM-Guided Generation of Safety-Critical Scenarios with Prior Knowledge
SMART: Scalable Multi-Agent Reasoning and Trajectory Planning in Dense Environments
REACT: Runtime-Enabled Active Collision-Avoidance Technique for Autonomous Driving
RiskNet: Interaction-Aware Risk Forecasting for Autonomous Driving in Long-Tail Scenarios
LEAD: Learning-Enhanced Adaptive Decision-Making for Autonomous Driving in Dynamic Environments
Understanding Driver Cognition and Decision-Making Behaviors in High-Risk Scenarios: A Drift Diffusion Perspective
Online Adaptive Platoon Control for Connected and Automated Vehicles via Physics Enhanced Residual Learning
SafeDrive: Knowledge-and Data-Driven Risk-Sensitive Decision-Making for Autonomous Vehicles with Large Language Models
Unveiling Uniform Shifted Power Law in Stochastic Human and Autonomous Driving Behavior
General Optimal Trajectory Planning: Enabling Autonomous Vehicles with the Principle of Least Action
CSDO: Enhancing Efficiency and Success in Large-Scale Multi-Vehicle Trajectory Planning
Attention-based Priority Learning for Limited Time Multi-Agent Path Finding
Deception for Advantage in Connected and Automated Vehicle Decision-Making Games
Human Decision-Making in High-Risk Driving Scenarios: A Cognitive Modeling Perspective
Online Physical Enhanced Residual Learning for Connected Autonomous Vehicles Platoon Centralized Control
Intelligent Adaptive Decision-Making for Autonomous Vehicles: A Learning-Enhanced Game-Theoretic Approach in Interactive Scenarios
Best Paper Award
Risk Generation and Identification of Driver-Vehicle-Road Microtraffic System
Best Paper Award
Probabilistic Situation Assessment for Intelligent Vehicles with Uncertain Trajectory Distribution
Towards The Unified Principles for Level 5 Autonomous Vehicles
Journal Cover Paper
An Integrated Architecture for Intelligence Evaluation of Automated Vehicles
Best Paper Award
A Probabilistic Risk Assessment Framework Considering Lane-Changing Behavior Interaction
Driving Risk Assessment Based on Naturalistic Driving Study and Driver Attitude Questionnaire Analysis
Selected Patents
Show selected patents
Modeling and Risk Identification Method for Driver-Vehicle-Road Microscopic Traffic System
A Method and Device for Proactive Response to Driver Risk and Decision-Making Under High-Risk Scenarios
Driving Risk Identification Model Calibration Method and System
Traffic Risk Assessment Method and Device
Intelligent Vehicle Safety Decision-Making Method Employing Driving Safety Field
Selected Talks & Presentations
Rethinking Safe Autonomy in the Long Tail
IEEE International Conference on Automation Science and Engineering (CASE 2026)
Aug 17, 2026
Safe and Trustworthy Autonomous Systems in Complex Environments: From Risk Perception to Decision-Making
Keynote Talk · FISITA Intelligent Safety Conference 2026
Jul 11, 2026
Safe and Adaptive Automated Driving Empowered by Data Intelligence and Generative AI
China Society of Automotive Engineers Forum
Dec 20, 2025
Towards Safe and Adaptive Urban Mobility with Multimodal Data and Generative AI
Onsite Job Talk · JTL Urban Mobility Lab, MIT
Aug 21, 2025
Towards Safe, Adaptive Autonomous Systems with Multimodal Data and Generative AI
Job Talk · Cho Chun Shik Graduate School of Mobility, KAIST
Aug 20, 2025
Towards Open-World Autonomy: Bridging Ground and Aerial Systems for Future Mobility
Onsite Job Talk · Department of Mechanical Engineering, KU Leuven
May 5, 2025
Learning-Enabled Human-Centered Autonomy for Trustworthy Deployment in Complex Environments
Job Talk · Oxford Robotics Institute
Apr 14, 2025
Probabilistic Risk Assessment and Sequential Decision-Making for Real-World Safety-Critical Systems
Institute for AI Industry Research, Tsinghua University
Nov 2023
Show additional talks & presentations
Towards Safe, Adaptive, Learning-based Control for Autonomous Systems
Onsite Job Talk · Department of Mechanical Engineering, KU Leuven
May 6, 2025
Uncertainty-Aware Decision-Making for Autonomous Driving in Driver-Vehicle-Road Coupled Environments
National Doctoral Academic Forum of the China SAE, Jilin University
Jun 2023
Towards Driver Risk Perception Mechanism-Driven Decision-Making for Intelligent Vehicles
Institute for AI Industry Research, Tsinghua University
Jun 2023
Towards High-Level Autonomous Driving of Vehicle Intelligent Safety Technology
Delft Center for Systems and Control, Delft University of Technology
Oct 2022
Risk-Aware Driver Model for Autonomous Vehicles based on the Principle of Least Action
Vehicle Engineering Graduate Academic Forum in Beijing Universities, Tsinghua University
May 2022
Probabilistic Driving Risk Assessment Considering Lane-Changing Behavior Interaction
Department of Cognitive Robotics, Delft University of Technology
Mar 2022
Human-Centered Risk Cognition in Autonomous Driving
639th Annual THU Doctoral Forum, Tsinghua University
Jun 2021
Path Planning for Vehicle Obstacle Avoidance Based on Collaborative Perception
558th Annual THU Doctoral Forum, Tsinghua University
May 2019
Teaching Service
MO.50000 Introduction to Mobility Systems
RO47015 Applied Experimental Methods: Human Factors
ME40150723 Intelligent and Connected Vehicle
EE30230783 Probability Theory and Stochastic Processes
Editorial, Organizing, and Membership
Associate editor
Editorial board member
Organizer
Guest editor
Committee membership
Peer Review
Journals
Conferences
Inside ACES Lab
Research platforms and lab life
Experimental systems that connect learning, perception, reasoning, and control in the physical world.
Hardware
Platforms for robotics, autonomous driving, and human-in-the-loop experiments.
Humanoid robotics
Unitree G1 Humanoid
A whole-body platform for studying balance, locomotion, and coordinated motion through imitation and reinforcement learning under physical constraints.
Legged robotics
Unitree Go2 Quadruped
A mobile testbed for learned locomotion, navigation, balance, and obstacle avoidance over uneven and changing terrain.
Autonomous driving
Autonomous Vehicle
An integrated platform for sensing, prediction, planning, and control, with an emphasis on safe behavior under uncertainty and incomplete observations.
Driving simulation
Driving Simulator
A controlled environment for investigating driving behavior, risk perception, decision-making, and human–automation interaction without real-world traffic exposure.
Gallery
Selected moments from research, conferences, and life together at ACES Lab.
