PhD Student · Computer Science · Purdue University
I am a Ph.D. student in the Department of Computer Science at Purdue University, co-advised by Prof. Aniket Bera and Prof. Lin Tan. My research focuses on physical AI and robot learning, particularly foundation models for robotics and reliable robotic systems. Previously, I worked on coding agents and code generation.
World Action Models, Vision-Language-Action Models, Robot Learning, and Safe Autonomy.
LLM Planners, Agentic AI, Reinforcement Learning, and Preference Optimization.
A corrective VLA policy that refines the action chunks proposed by a frozen flow-based VLA before execution, trained with SFT followed by flow-based correction preference optimization. Consistent performance gains across three VLA backbones π0.5, FLOWER VLA, and Xiaomi-Robotics-0.
NL-to-LTL Translation, LTL-constrained decoding, and domain-specific fine-tuning make LLM planners follow complex and conditional user instructions, improving safety rate by 10.8% and plan efficiency by 19.8% on robot manipulation and drone navigation tasks.
A study of 1,200 Claude Code and Mini-SWE-Agent trajectories on SWE-bench that identifies three recurring cost-inefficient behaviors, then shows developer-designed skills cut task cost by up to 41.73%, roughly twice the gain of agent-synthesized skills.
How Effective Are Neural Networks for Fixing Security Vulnerabilities. Yi Wu, Nan Jiang, Hung Viet Pham, Thibaud Lutellier, Jordan Davis, Lin Tan, Petr Babkin, Sameena Shah.
Anti-patterns for Java Automated Program Repair Tools. Yi Wu. Won 1st Place.
Advisors: Prof. Aniket Bera and Prof. Lin Tan
West Lafayette, IN, USA
Shenzhen, China
Purdue CS News: SELP named a Best Paper Award Finalist at ICRA 2025.
SUSTech CS News: First place in the ACM Student Research Competition at ASE 2020.