Hello! I am Yinan Zheng. I am a 3rd-year PhD candidate at AIR, Tsinghua University, advised by Prof. Xianyuan Zhan, Prof. Shengbo Eben Li and Prof. Jingjing Liu.
My research focuses on advancing AI-powered solutions for superhuman-level and safe real-world decision-making and LLMs. Currently, I am working on:
- (IL/RL+X) Advancing IL/RL for scalable, efficient, and superhuman performance in decision-making and LLMs.
- (GenAI+X) Leveraging the power of generative models to enhance planning, control in robotics and autonomous driving.
- (SafeAI+X) Ensuring the reliability of learning-based decision-making systems and the alignment of LLMs.
I am open to collaboration, feel free to reach me out!
Some links: Github / Twitter / Google Scholar / zhengyn23@mails.tsinghua.edu.cn
News
- RefectDrive, a discrete diffusion-based VLA model with safety-aware reflection, is now available on arXiv.
- Two papers on autonomous driving (Flow-Planner) and zeroshot-rl (BREEZE) are accepted to NeurIPS 2025.
- One paper (LBP) on efficient latent planning is accepted to ICML 2025.
- One paper (UniAct) on cross-embodiment universal actions is accepted to CVPR 2025.
- πDiffusion-Planner is selected as oral presentation at ICLR 2025.
- Two papers on fast post-train (PSEC) and autonomous driving (Diffusion-Planner) are accepted to ICLR 2025.
- πIVM and DecisionNCE are selected as Outstanding Paper at MFM-EAI workshop @ ICML 2024.
- One paper (IVM) on embodied foundation multimodal models is accepted to NeurIPS 2024.
- One paper (DecisionNCE) on embodied multimodal representations is accepted to ICML 2024.
- πOne paper (FISOR) on safe offline rl is accepted to ICLR 2024.
- One paper (OMIGA) on offline multi-agent rl is accepted to NeurIPS 2023.
Publications (* marks equal contribution)
- Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving Arxiv 2025 Paper | Code
- Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling NeurIPS 2025 2025 Paper | Code
-
Diffusion-Based Planning for Autonomous Driving with Flexible Guidance
ICLR 2025 (Oral, Top 2%) 2025 Paper | Code | Page
- DecisionNCE: Embodied Multimodal Representations via Implicit Preference Learning ICML 2024 (Outstanding Paper @ ICML 2024 MFM-EAI Workshop) 2024 Paper | Code | Page
-
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
ICLR 2024 2024 Paper | Code | Page
- Diffusion-Based Planning for Autonomous Driving with Flexible Guidance ICLR 2025 (Oral, Top 2%) 2025 Paper | Code | Page
- Skill Expansion and Composition in Parameter Space ICLR 2025 2025 Paper | Code | Page | Dataset | Model
- Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model ICLR 2024 2024 Paper | Code | Page
- Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving Arxiv 2025 Paper | Code
- Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling NeurIPS 2025 2025 Paper | Code
- Towards Robust Zero-Shot Reinforcement Learning NeurIPS 2025 2025
- Efficient Robotic Policy Learning via Latent Space Backward Planning ICML 2025 2025 Paper | Code | Page
- Universal Actions for Enhanced Embodied Foundation Models CVPR 2025 2025 Paper | Code | Page
- Diffusion-Based Planning for Autonomous Driving with Flexible Guidance ICLR 2025 (Oral, Top 2%) 2025 Paper | Code | Page
- Skill Expansion and Composition in Parameter Space ICLR 2025 2025 Paper | Code | Page | Dataset | Model
- Instruction Guided Visual Masking NeurIPS 2024 (Outstanding Paper @ ICML 2024 MFM-EAI Workshop) 2024 Paper | Code | Page | Dataset | Model
- DecisionNCE: Embodied Multimodal Representations via Implicit Preference Learning ICML 2024 (Outstanding Paper @ ICML 2024 MFM-EAI Workshop) 2024 Paper | Code | Page
- Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model ICLR 2024 2024 Paper | Code | Page
- Offline Multi-Agent Reinforcement Learning with Implicit Global-to-Local Value Regularization NeurIPS 2023 2023 Paper | Code
Professional Services
Reviewer for ICLR, ICML, NeurIPS (Top Reviewer)
