vision language action research

RT-2

RT-2 adapted web-scale vision-language representations into a vision-language-action model for robot control.