01
Vision-Language-Action Models
Status monitoring, error recovery, dynamic reasoning, and test-time compute scaling for embodied agents.
University of Sydney ยท Ph.D. Researcher
Wenhao Li builds AI systems that act, verify, and improve.
I am a Ph.D. student in Computer Science at the University of Sydney, advised by Prof. Chang Xu. My work focuses on Vision-Language-Action models for embodied intelligence, multimodal LLM reliability, efficient diffusion models, and autonomous research agents.
Research Focus
My current research connects embodied agents, multimodal reasoning, model self-monitoring, and efficient generative modeling.
02
Dynamic Manipulation
Motion-aware VLA systems for conveyor-belt and other latency-sensitive manipulation settings.
03
Reliable Multimodal LLMs
Identifying, isolating, and purging hallucination components through self-evolving distillation.
04
Efficient Diffusion & Agents
Architecture search, progressive scaling, and research-agent systems that explore and accumulate experience.
Selected Work
A snapshot of recent first-author work across embodied AI, multimodal reliability, and efficient generation.
Sentinel-VLA: A Metacognitive VLA Model with Active Status Monitoring for Dynamic Reasoning and Error Recovery
Equips VLA models with active execution-state monitoring and recovery behavior for robust embodied tasks.
VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model
Uses an action critic to allocate deliberation at test time when a task requires more careful reasoning.
Identify, Isolate, and Purge: Mitigating Hallucinations in LVLMs via Self-Evolving Distillation
Analyzes and removes hallucination-related knowledge components to improve multimodal model reliability.
DiffNAS: Bootstrapping Diffusion Models by Prompting for Better Architectures
Uses LLM-guided architecture search to discover stronger diffusion backbones under practical budgets.
Experience
Research training across academia, robotics, and industrial AI labs.
Ph.D. in Computer Science, University of Sydney
Advised by Prof. Chang Xu, working on VLA, embodied intelligence, multimodal models, and generative AI.
Embodied LLM Algorithm Research Intern, AgiBot Embodied Research Center
Research on VLA state monitoring, error handling, and test-time scaling.
Full-time Algorithm Researcher, SenseTime Research
Worked on multimodal hallucination mitigation and lightweight diffusion-based image generation.
B.S. in Software Engineering, Beihang University
Graduated with an average score of 89/100 and multiple first-class innovation scholarships.