PI-0.5: ROBOTS IN THE WILD

Summary from the TLDR AI newsletter

The Physical Intelligence team has tested its house cleaning robot in novel unseen environments and found it performs quite well. The team achieved this with a mixture of VLM training and action tokenization.