Wenhao Lu

PhD researcher in robot learning and AI

I am a PhD researcher in robot learning and AI. My work focuses on reinforcement learning for robotic manipulation and on interpretability methods that make RL agents easier to analyze, explain, and trust.

research themes

RL-Driven Robot Learning

I study reinforcement learning methods for robotic manipulation and embodied control, with an emphasis on reward design, obstacle-aware behavior, and learning from interaction.

Interpretable RL Agents

I develop methods and evaluations for understanding why RL agents act, including reward decomposition, causal state distillation, action coherence, and trajectory-level explanations.

Language Models for RL Explanations

I use language models as tools for querying, summarizing, and stress testing explanations of RL agents, while keeping the learned RL policy and its behavior as the central object of study.

  1. ICDL 2023
    A Closer Look at Reward Decomposition for High-Level Robotic Explanations
    Wenhao Lu ,  Xufeng Zhao ,  Sven Magg , and 3 more authors
    In IEEE International Conference on Development and Learning (ICDL 2023) , Nov 2023
    Oral presentation
    Abs arXiv Bib
  2. RD_flow_overview.jpg
    Causal State Distillation for Explainable Reinforcement Learning
    Wenhao Lu ,  Xufeng Zhao ,  Thilo Fryen , and 4 more authors
    In 3rd Conference on Causal Learning and Reasoning (CLeaR 2024) , Apr 2024
    Oral presentation
    Abs arXiv Bib Website
  3. COLING 2024
    Enhancing Zero-Shot Chain-of-Thought Reasoning in Large Language Models through Logic
    Xufeng Zhao ,  Mengdi Li ,  Wenhao Lu , and 4 more authors
    In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , May 2024
    Oral presentation
    Bib
  4. ICANN 2024
    Details Make a Difference: Object State-Sensitive Neurorobotic Task Planning
    Xiaowen Sun ,  Xufeng Zhao ,  Jae Hee Lee , and 3 more authors
    In Artificial Neural Networks and Machine Learning - ICANN 2024 , May 2024
    arXiv Bib
  5. Humanoids 2024
    Large Language Models for Orchestrating Bimanual Robots
    Kun Chu ,  Xufeng Zhao ,  Cornelius Weber , and 3 more authors
    In 23rd IEEE-RAS International Conference on Humanoid Robots (Humanoids 2024) , May 2024
    arXiv Bib
  6. TMLR 2025
    Mental Modeling of Reinforcement Learning Agents by Language Models
    Wenhao Lu ,  Xufeng Zhao ,  Josua Spisak , and 2 more authors
    Transactions on Machine Learning Research, May 2025
    Also presented at the 18th European Workshop on Reinforcement Learning
    Abs arXiv Bib Website
  7. ACL Findings 2026
    Curriculum-RLAIF: Curriculum Alignment with Reinforcement Learning from AI Feedback
    Jiaye Lin ,  Mengdi Li ,  Xufeng Zhao , and 4 more authors
    May 2026
    ACL 2026 Findings
    arXiv Bib

selected projects

RL-Driven Robotic Manipulation

Reinforcement learning for robotic manipulation, including reward design, obstacle-aware behavior, and simulation-based evaluation.

robot learning reinforcement learning manipulation

Causal and Language-Based RL Interpretability

Analyses of RL agents using causal state distillation and language-model mental modeling to inspect learned behavior from trajectories.

RL interpretability causal analysis mental modeling

research vision

My long-term goal is to develop RL-driven robotic systems whose behavior can be improved, inspected, and explained. I am especially interested in the connection between reinforcement learning, embodied interaction, and interpretability: agents should not only complete tasks, but also expose reliable evidence about why they act.

technical capabilities

Reinforcement Learning

PPO reward shaping reward decomposition trajectory-level evaluation

Robot Learning

robotic manipulation simulation-based learning obstacle-aware behavior embodied control

Interpretability

causal state distillation high-level robotic explanations mental modeling of RL agents action coherence analysis

Engineering Tools

PyTorch robotics simulation experiment analysis TODO: verify additional tools before listing them here