--
You received this message because you are subscribed to the Google Groups "RSSC-List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rssc-list+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/rssc-list/CAPjT4OfyKLyuKK5djK-yDXdL6frZTXBSoJbADx5kr9Ka4zyuHw%40mail.gmail.com.
To view this discussion visit https://groups.google.com/d/msgid/rssc-list/CAOgm0y51S6XJf3WHTzmgiJgmi92kd6P9Acr4uJGPkT%3DC9roDdw%40mail.gmail.com.
Gentlemen, I think there's some fundamental mischaracterization happening in this thread that's worth straightening out.
These are not pre-scripted motions. There is no choreography file, no motion-capture replay, no CNC-style waypoint program producing what you're seeing. If you've ever tried to hand-code a humanoid balancing on one leg while transitioning between dynamic poses, you'd appreciate why that framing doesn't hold up.
What you're actually looking at is a cutting-edge demonstration of reinforcement learning policies that are robust to real-world physical uncertainty — balance perturbations, surface variation, actuator imprecision. These behaviors are almost certainly trained in simulation (most likely NVIDIA Isaac Sim / Isaac Lab or a comparable physics-based RL pipeline) and then transferred to physical hardware. The robots aren't following a script; they're executing learned policies that dynamically adapt to maintain balance and hit target poses. That's a genuinely hard problem, and dismissing it as "just a CNC machine in a human shape" misses the point entirely.
For context, there are two major branches of physical robot AI converging right now:
Low-level motor policy learning via RL — walking, running, dancing, recovering from pushes, backflips. This is where many Chinese robotics companies (Unitree, UBTECH, etc.) have been making dramatic progress, and it's exactly what this video demonstrates.
Higher-level manipulation via behavioral cloning and RL — "find the can, pick it up, hand it to me." Companies like Figure, Boston Dynamucsm 1x, hobbyists like Me!, and various university labs are pushing this frontier.
These aren't competing approaches — they're complementary layers of the same stack, and they're converging fast.
Now — can these robots fetch you a beer from the fridge? Not reliably. Not yet. But "it can't do everything therefore it's nothing special" is a strange place to plant your flag. By that logic, the first airplane was unimpressive because it couldn't fly transatlantic.
And Chris — I'd pay real money to see a CNC machine do a backflip. :-)
The future is closer than you think -- I wouldn't intentionally ignore the signs of it's approach. :-)