Robotics
/robotics1794
a place to discuss robotics, how to build them, papers, tutorials, the consequences, hands on demos and more
They can also be used to carry out initial quality checks of parts. It’s not exactly the stuff of sci-fi movies, but it’s a glimpse of what the future of factory work might look like.
https://www.carscoops.com/2025/03/mercedes-humanoid-robots/
https://www.carscoops.com/2025/03/mercedes-humanoid-robots/
just started a new pcb for the tumbller cryptobot in RPi Hat form factor
wut? humanoid half marathon in beijing
https://english.beijing.gov.cn/latest/news/202503/t20250307_4028142.html
https://english.beijing.gov.cn/latest/news/202503/t20250307_4028142.html
Robot Delivery day!
If they were looking for volunteers to test this at home, would you sign up?
https://techcrunch.com/2025/02/27/figure-will-start-alpha-testing-its-humanoid-robot-in-the-home-in-2025/
https://techcrunch.com/2025/02/27/figure-will-start-alpha-testing-its-humanoid-robot-in-the-home-in-2025/
Humanoids might just be the most versatile form factor, not skeuomorphism.
https://warpcast.com/s5eeo/0xbdbb4033
https://warpcast.com/s5eeo/0xbdbb4033
Been testing this MyCobot robot arm. It has 6 degrees of freedom(!), a Python API, and runs on two ESP32s, so plenty of interfacing options.
But the calibration software doesn’t work (Mac issue?), and the docs are a mess.
Most importantly, it’s highly suicidal. It doesn't come with anything to secure it, so it just flings itself around. For now, I’m holding it down.
Next step: Design and 3D print a way to secure this thing before it takes me out.
But the calibration software doesn’t work (Mac issue?), and the docs are a mess.
Most importantly, it’s highly suicidal. It doesn't come with anything to secure it, so it just flings itself around. For now, I’m holding it down.
Next step: Design and 3D print a way to secure this thing before it takes me out.
What seems impossible today could soon become common.
https://www.youtube.com/embed/uVcBa6NXAbk?si=Oc2i01i2PzSEHHpJ"
https://www.youtube.com/embed/uVcBa6NXAbk?si=Oc2i01i2PzSEHHpJ"
In the beginning of automobiles era (1885-1920s) low key it’s easy to forget how cars had no dedicated infrastructure. Early on, drivers would take many spare tires with them because you would get many flats. The roads weren't developed, they were all dirt roads, shared with horses and carriages. No gasoline stations, no safety rules, nothing really. Just a weird fun vehicle - now look at today, and where we are.
In this context -- I am actually excited about Humanoids as a product category. It's early, and most of what is out there today is shit (and the future that it extrapolates)
We can do better
In this context -- I am actually excited about Humanoids as a product category. It's early, and most of what is out there today is shit (and the future that it extrapolates)
We can do better
Part of the rise of the current batch of humanoids have worked well is because "Behavior Cloning" has made RL a more solvable problem that stumped robotics for years. That's why end-to-end policies that run directly from images is exploding. What that means:
Why it's exciting:
- Classical robotics: traditionally relied on explicitly programming, the controls, planning, models etc define state-action mappings, tune params, and build for edge cases. Does not generalize well
- Requires little explicit programming -- you just need to collect data of an expert performing a task, then train a model to copy it. It's supervised by nature. Then doing new tasks is as simple as collecting more data. No need to constantly build and support for unknown edge cases
https://www.imgeorgiev.com/2025-01-31-why-bc-not-rl/
Why it's exciting:
- Classical robotics: traditionally relied on explicitly programming, the controls, planning, models etc define state-action mappings, tune params, and build for edge cases. Does not generalize well
- Requires little explicit programming -- you just need to collect data of an expert performing a task, then train a model to copy it. It's supervised by nature. Then doing new tasks is as simple as collecting more data. No need to constantly build and support for unknown edge cases
https://www.imgeorgiev.com/2025-01-31-why-bc-not-rl/
I think there would be lots of custom robotics opportunities for me in India
https://www.bbc.com/news/world-asia-india-64740853.amp
https://www.bbc.com/news/world-asia-india-64740853.amp
The agent → bot reverse transition will be rapid.
π₀ is open sourced - now all you need is a setup
https://x.com/physical_int/status/1886822689157079077?s=46
https://x.com/physical_int/status/1886822689157079077?s=46
"the...future"
chinese companies also ship great lidars besides the usual things in news
https://x.com/robosenselidar/status/1876629869687443840
https://x.com/robosenselidar/status/1876629869687443840
They run now (Unitree G1)
Better looking gait, looks more natural
https://www.youtube.com/watch?v=CIkdq7Zf4Zw
Better looking gait, looks more natural
https://www.youtube.com/watch?v=CIkdq7Zf4Zw