ScaleVLA ties a 552B VLM backbone to a robot action model
With 10 demonstrations it beat a 2B backbone trained on 50, and LIBERO-PRO success was 45.1% versus at most 31.3%.
1 stories, newest first · one card per story
With 10 demonstrations it beat a 2B backbone trained on 50, and LIBERO-PRO success was 45.1% versus at most 31.3%.
Today 959 readers · 1.2K views · 163 reads · all time 202K views