Physical Agentic AIAction GenerationRobotics
I believe actions should be generated by models that deeply learn the action modality itself, rather than relying on priors from video or vision-language models. We need these action models, along with benchmarks to evaluate them.
With strong action models in place, the next question is how to use them effectively. This is where my interests in Action Generation and Physical Agentic AI come together.
My goal is to combine my research on action generation with the Physical Agentic AI systems I develop, building a zero-shot system that works reliably in the physical world and makes our everyday lives easier.