CLAP: Converting Vision-Language Models into Vision-Language-Action Models via Language-Prompted Actions
Converts any pretrained vision-language model into a vision-language-action model with zero architectural changes by prepending natural-language action descriptions; reaches 90.8% success on LIBERO with a 2B model after less than 6 hours of post-training.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.