Hey, I’m a Ph.D. student at the Trustworthy Autonomous Systems Laboratory (TASL), Georgia Institute of Technology. It is my honor to be advised by Prof. Jiachen Li.

My research focuses on reliable, human-centered autonomy, with an emphasis on multi-agent learning and decision-making under uncertainty. Building on efficient multi-agent communication, I develop robust algorithms for collaborative perception, prediction, and planning, enabling autonomous systems to coordinate effectively and align with human intent and preferences in real-world environments.

🔥 News

• 2026.09 LIBERO-PeRM accepted to NeurIPS 2026 Datasets and Benchmarks Track
• 2026.09 We will host the Human-Centered Robot Learning and Interaction (HumanRLI) workshop at CoRL 2026 (Austin, TX, Nov 12)
• 2026.09 Moved to Georgia Tech with the TASL group.
• 2026.09 CoopUQ and HiRe-MoMa accepted to CoRL 2026
• 2026.06 Started my research internship at Bosch USA, working on Vision–Language–Action models for long-horizon navigation.
• 2026.06 NavTrust accepted to IROS 2026
• 2026.02 Drive My Way accepted to CVPR 2026

🎯 Research Interests

• Multi-Agent Systems: communication-efficient cooperative perception, motion prediction, and decision-making
• Autonomous Driving: Vision–Language–Action (VLA) models, RL post-training
• Human-Robot Interaction: human-guided policy learning, preference alignment, social navigation

📝 Selected Publications

LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation

Zhixu Li, Keqian Tang, Litian Gong, Jingyu Yao, Wenqian Zhang, Tianze Xu, Zehao Wang, Yuping Wang, Junge Zhang, Jiachen Li^‡

NeurIPS 2026 Conference on Neural Information Processing Systems, Datasets and Benchmarks Track

• A benchmark for personalized robotic manipulation, treating user preference as a controllable variable alongside tasks and layouts.
• A procedural pipeline generates paired demonstrations with and without preferences, plus preference-satisfaction metrics and subtask-level metadata.
• Five task suites (312 tasks) probe zero-shot personalization, learning ability and efficiency, generalization, and preference composition and conflict.
• Current VLA policies struggle with zero-shot personalization; SFT makes preferences learnable, but generalization and conflict resolution remain difficult.

CoopUQ: Robust Multi-Robot Social Navigation via Uncertainty-Aware Cooperative Forecasting

Zehao Wang, Jianpeng Yao, Yuping Wang, Yiping Li, Jiaqi Li, Zhixu Li, Kanghoon Lee, Jinfeng Wang, Cong Liu, Zhaowei Tan, Jinkyoo Park, Jiachen Li^‡

CoRL 2026 Conference on Robot Learning

• Cooperative occupancy forecasting through cross-robot spatiotemporal feature fusion.
• Constrained multi-agent RL penalizes regions predicted occupied or highly uncertain, adapting risk sensitivity under distribution shift.
• Outperforms baselines in prediction accuracy and safety on in-distribution and OOD Unity3D scenes, and is validated on real robots.

Hierarchical Residual Policy Learning for Real-World Mobile Manipulation with Sparse Human Guidance

Zhefei Gong, Hefeifei Jiang, Zehao Wang, Jiachen Li^‡

CoRL 2026 Conference on Robot Learning

• A policy-agnostic residual policy fine-tunes a frozen base policy for real-world mobile manipulation.
• The residual splits into base and arm heads with unidirectional conditioning, avoiding gradient interference.
• Expert-guided value alignment (under 20% intervention) lifts TIAGo++ performance in ~30 min of on-robot interaction.

Drive My Way: Preference Alignment of Vision–Language–Action Models for Personalized Driving

Zehao Wang, Huaide Jiang, Shuaiwu Dong, Yuping Wang, Hang Qiu, Jiachen Li^‡

CVPR 2026 IEEE/CVF Conference on Computer Vision and Pattern Recognition

📄 arXiv 🌐 Project 💻 Code

• A VLA model aligned to both long-term driving patterns and real-time preference instructions.
• GRPO post-training with a residual decoder and adaptive style rewards.
• Validated through closed-loop evaluations and user studies.

NavTrust: Benchmarking Trustworthiness for Embodied Navigation

Huaide Jiang^*, Yash Chaudhary^*, Yuping Wang, Zehao Wang, Raghav Sharma, Manan Mehta, Yang Zhou, Lichao Sun, Zhiwen Fan, Zhengzhong Tu, Jiachen Li^‡

IROS 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems

📄 arXiv 🌐 Project 💻 Code

• A benchmark for measuring the trustworthiness of embodied navigation agents.

CMP: Cooperative Motion Prediction with Multi-Agent Communication

Zehao Wang^*, Yuping Wang^*, Zhuoyuan Wu^*, Hengbo Ma, Zhaowei Li, Hang Qiu^‡, Jiachen Li^‡

RA-L 2025 IEEE Robotics and Automation Letters

📄 arXiv 🌐 Project 💻 Code 🎥 Video

• Latency-robust cooperative motion prediction from information shared across multiple CAVs.
• Unifies information sharing across both the perception and prediction modules.
• Extensive experiments and ablations on OPV2V and V2V4Real.

🧑‍🏫 Academic Service

• Workshop Organizer: Human-Centered Robot Learning and Interaction (HumanRLI), CoRL 2026
• Conference Reviewer: CVPR, ICCV, CoRL, ICRA, IROS
• Journal Reviewer: RA-L

🎖 Selected Awards

• Dean’s Distinguished Fellowship, UC Riverside (2023)
• Merit Scholarship (2021)
• Outstanding Graduate (2020)
• National Scholarship (2019)

📖 Education

• 2026.09 – now
  Ph.D. in Machine Learning Georgia Institute of Technology
  
• 2023.09 – 2026.08
  Ph.D. Candidate in Computer Science University of California, Riverside
  
• 2021.09 – 2023.05
  M.S. in Computer Science New York University
  
• 2016.09 – 2020.06
  B.S. in Software Engineering Sun Yat-sen University