An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving
IEEE International Conference on Intelligent Transportation Systems (ITSC) (ITSC 2026)
Yun Li
PhD candidate · JSPS DC2 Research Fellow · Student Researcher at TIER IV
Generative AI · World models · Autonomous driving
Tsukada Lab, The University of Tokyo
I am a PhD candidate at Tsukada Lab, Graduate School of Information Science and Technology, The University of Tokyo, graduating September 2026. JSPS picked up my DC2 proposal in 2026 (KAKENHI 26KJ0799).
My work is on generative models for autonomous driving: how to make VLA driving policies safe, and how to make diffusion trajectory planners fast enough to deploy. The thesis is Safe and Efficient Generative Models for Autonomous Driving: From Preference-Aligned VLA to Real-Time Diffusion Planning. Two lines of work:
- Preference alignment for VLA driving. PrefDrive at IV 2025 was the first DPO-based driving policy I built; Multi-PrefDrive at IROS 2025 extended it to Plackett–Luce ranking; a Japanese short paper covers the same theme at JSAI 2026.
- Real-time diffusion planning. The Modular Diffusion Benchmark at ITSC 2026 in Naples breaks a monolithic ONNX diffusion planner into swappable components; the ICML 2026 RLxF workshop paper proves a no-go for bit-exact safety rewards and offers an ODD-adaptive shared/expert LoRA that gets around it.
Education
- PhD candidate, Information Science and Technology, The University of Tokyo (2023–2026)
- M.Eng., Information & Communications Engineering, Tokyo Institute of Technology (2021–2023) GPA 3.57/4.0
- B.Eng., Vehicle Engineering, Jilin University (2017–2021) exchange at Nagoya University
- Research Assistant, Auto Simulation & Control State Key Lab, Jilin University (2020–2022)
Experience
- Research Intern, Huawei 2012 Labs, Shenzhen (Jul–Nov 2025)
- Visiting Scholar, Nakahira Lab, CMU (Apr–May 2025)
- Student Researcher, TIER IV (Apr 2024 – present)
- Research Intern, NEC Data Science Research Labs (Aug–Sep 2022)
Invited talks
- “Preference-Guided LLMs for Driving Assistance”, SPADE Workshop @ IV 2025, Cluj-Napoca (talk + panel).
- “Foundation Models for Autonomous Driving”, FMAD Workshop @ ITSC 2024, Edmonton.
- “Large Language and Vision Models for Autonomous Driving”, LLVM-AD Workshop @ ITSC 2024, Edmonton.
- “Vision Language Model based Human-Centered Autonomous Driving”, VLM-AD Workshop @ ITSC 2024, Edmonton.
Patent
- Yun Li, R. Zhao, Z. Gao. “Conflict-free cooperation method for self-driving vehicles at intersections based on deep reinforcement learning.” China Patent CN115457782A, 2023.
Featured video
Latest News
- Attended the JSPS Fellowship Friendship Meeting 2026
- PhD preliminary defense at Tsukada Lab
- JSPS DC2 fellowship (KAKENHI 26KJ0799) awarded
- Modular Diffusion Benchmark accepted at ITSC 2026
- Multi-PrefDrive accepted at IROS 2025
Publications
2026
Learning Diffusion Planners from World Feedback: A No-Go Result on Bit-Exact Safety Rewards and an ODD-Adaptive Shared/Expert Decomposition
Workshop on Reinforcement Learning from World Feedback (RLxF), ICML 2026 (ICML 2026 RLxF Workshop)
正則化付き多重選好学習による自動運転 VLA モデルの安全制約アライメント
JSAI 2026 (人工知能学会全国大会) (JSAI 2026)
2025
Multi-PrefDrive: Optimizing Large Language Models for Autonomous Driving Through Multi-Preference Tuning
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IROS 2025)
PrefDrive: Enhancing Autonomous Driving through Preference-Guided Large Language Models
IEEE Intelligent Vehicles Symposium (IV) (IV 2025)
Centralized Cooperation for Connected Autonomous Vehicles at Intersections by Safe Deep Reinforcement Learning
IEEE Transactions on Mobile Computing (TMC) (IEEE TMC [IF 7.9, CCF A])
Autonomous Intersection Management via Prior-Enhanced Multi-Agent Constrained Decision Transformer
IEEE Transactions on Intelligent Transportation Systems (T-ITS) (IEEE T-ITS [IF 8.5, Q1])
2024
Large Language Models for Human-like Autonomous Driving Decision Making: A Survey
IEEE International Conference on Intelligent Transportation Systems (ITSC) (IEEE ITSC 2024)
Vehicle Collision Warning Method at Intersection Based on V2I Communication
Journal of Jilin University (J. Jilin Univ.)
Sce2DriveX: A Generalized MLLM Framework for Scene-to-Drive Learning
IEEE Robotics and Automation Letters (RA-L) (IEEE RA-L [IF 5.2, Q1])
Safe Multi-Agent Deep Reinforcement Learning for the Management of Autonomous Connected Vehicles at Future Intersections
IEEE Transactions on Parallel and Distributed Systems (TPDS) (IEEE TPDS [IF 5.6, CCF A])
A Survey on Recent Advancements in Autonomous Driving Using Deep Reinforcement Learning: Applications, Challenges, and Solutions
IEEE Transactions on Intelligent Transportation Systems (T-ITS) (IEEE T-ITS [IF 7.9, CCF A])
2023
Reinforcement Learning-Based Cognitive Radio Transmission Scheduling in Vehicular Systems
IEEE 97th Vehicular Technology Conference (VTC-Spring) (IEEE VTC-Spring 2023)
Multi-Agent Constrained Policy Optimization for Conflict-Free Management of Connected Autonomous Vehicles at Unsignalized Intersections
IEEE Transactions on Intelligent Transportation Systems (T-ITS) (IEEE T-ITS [IF 7.9, CCF A])
Centralized Cooperative Control for Autonomous Vehicles at Unsignalized All-Directional Intersections: A Multi-Agent Projection-Based Constrained Policy Optimization Approach
Expert Systems with Applications (ESWA [IF 8.5, Q1])