Really insightful presentation! The two-level framework for training humanoid robots using Vision-Language-Action models combined with low-level manipulation skills seems like a promising pathway to creating more versatile robots. I’m particularly intrigued by the potential of leveraging human video data to teach robots precise actions based on natural language instructions-it’s exciting to think about how this can enable robots to handle real-world tasks. The integration of Sim2Real is also a key factor for bridging the gap between simulation and real-world environments. However, I can see that the challenge of limited data for training more complex tasks is something that will need continued focus. Looking forward to seeing how this framework evolves in the future!
@33:00 re unfeasible lower body motion - the boston dynamics bipeds are already capable of inertially significant motion: backflips, catch & throw heavy bags &c
Really insightful presentation! The two-level framework for training humanoid robots using Vision-Language-Action models combined with low-level manipulation skills seems like a promising pathway to creating more versatile robots. I’m particularly intrigued by the potential of leveraging human video data to teach robots precise actions based on natural language instructions-it’s exciting to think about how this can enable robots to handle real-world tasks.
The integration of Sim2Real is also a key factor for bridging the gap between simulation and real-world environments. However, I can see that the challenge of limited data for training more complex tasks is something that will need continued focus. Looking forward to seeing how this framework evolves in the future!
@33:00 re unfeasible lower body motion - the boston dynamics bipeds are already capable of inertially significant motion: backflips, catch & throw heavy bags &c
Interesting 🤖
Thanks, I will make humanoids that will be called Android 17 and 18.
#ich können......
#@mostafasaadinasab