Kevin Black
Kevin Black is a researcher at Physical Intelligence and a PhD student at Berkeley AI, where he is advised by Sergey Levine. He is a lead author on the π0 and π0.5 vision-language-action flow models for general robot control, as well as the Octo open-source generalist robot policy. His research spans diffusion models, reinforcement learning, and robotic manipulation, with prior work including training diffusion models with reinforcement learning at Stanford and UC Berkeley.
“Stage 1 performs SFT on a task-specific dataset to obtain a base policy from a flow-matching VLA backbone [π0, π0.5]”
Source→“Kevin Black et al. (Physical Intelligence), Physical Intelligence, Lead authors of PI0 and PI0.5, the models VLA-ATTC is built on top of and benchmarked against. Their work is the de facto SOTA baseline for VLA performance.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.