Great work from @Machenta_@Ed__Johns !
Opening this water bottle with one single demonstration🤯
I was told it will be able to open other bottles with the same mechanism. What about longer horizon tasks? How well does it memorizes🤨
#CoRL @ Austin is fun! Industry and academic robotics research are blurring fast.
In Physical AI, deployment is the research. You can't deliver value without understanding real use cases, and you can't understand them without shipping.
@RussTedrake's industrial insights were great. But consumer and service are different: speed matters less, human-robot interaction matters more. The goal isn't replacing the whole workflow. It's designing robots to cowork with people, so we can see where they fail and fix it.
Real deployment is far messier than any lab benchmark. Can real-world RL close the gap, or is that wishful thinking?
@yukez Hi @yukez, we are hosting “Robotics Happy Hour” on Monday, could you add our event link to the CoRL website? Thank you!🙏🏼 https://t.co/7SjNWz0LHr
We will be co-hosting a CoRL after hours event on November 9th with @GentoRobotics . Looking forward to chat with everyone in person! https://t.co/oZBoYXokKI
Reasoning, spatial perception and even motor control (across different embodiment) is something frontier labs have advantages over robotics startups up. But how to do it reliably, at an inference level, perceiving the environment whilst adjusting is output? We have higher tolerance when it stays in a virtual world, but much less so in reality. Therefore I would say we are still not there yet, and both team have their respective strengths. One perhaps tackle the problem from top down, and the other from bottom up( through deployment know how’s).
Really enjoy playing with Poke’s RoboAPI in our lab. The whole system is plug-and-play. I’ll definitely stay tuned and look forward to their future updates.
Why out of the box capability is so important to us? The barrier for wider range robotics adaptations isn’t on hardware price along, but also what it can do for anyone without prior robotics background. This is what we work at Poke every day.
Log in to our platform and try it yourself!
Don't trust demos. Try it yourself.
Introducing RoboAPI, powered by Poke-Aurora, our world-action model.
Give compatible robots natural-language instructions through a simple API. No task-specific training.
More information: https://t.co/BHzTOo4I8u
@AaronWeiHuang Congrats on the work! Interesting to see that dynamic cup stacking shows history matters beyond long-horizon tasks, also memory and real-time can coexist in WAM.
@theSamPadilla fuddling around the concepts, giving false expectations to wider commercial sectors really jeopardizes the robotics community as a whole. aside from story telling, funding pressures, we need to to face the existing limitations...
@Z1hanW super cool work! Using only 4 seed videos is crazy… curious to know how the team decided on the variations of generated videos. More diversity on object forms? Heights?