Video yükleniyor...
Video Yüklenemedi
Atlas autonomously moving engine covers between containers and a mobile dolly. It uses an ML vision model to detect and localize environmental fixtures and individual bins. It uses a specialized grasping policy and continuously estimates the objects' states to achieve the task.
188,133 görüntüleme • 1 yıl önce •via X (Twitter)
10 Yorum

Vision model appears kinda primitive, or maybe that’s just the visual diagnostic output of a few NN layers. Seems (?) that the NN was trained on that particular shape. Impressive walking and body movement. Swivel head seems kinda silly when they could just have a rearward facing camera in the head like Optimus? Probably only Tesla has mastered fusing many cameras into a stitched together scene. They got that for free from their car FSD. Claw hands still seem to be limiting. Nice to see that this is end to end AI in some form, congrats BD! That’s a big change from their previous bots. All these different humanoid bots each have their strengths and weaknesses compared to each other. Still a few years away from any one company getting it all together for something generalizably useful.

The visualization could be a separate point cloud map to provide precise input for tasks and even be used for training. They may not yet have a fancy visual representation of the vision model. Tesla showed something similar from Optimus's PoV. Head turning seems unnecessary but probably helps human coworkers understand the robot's intentions more clearly. The waist and hip rotation make pivoting super efficient.

Lol, the difference in your tweets for different companies hints at a pretty big bias 😉

Interesting neck and limb idea to not limit it to the regular human ranges of motion.

After knowing how important the hands are, these robot video test seems way too simple and pretty pointless.

Specialized end-effectors make sense for specialized applications. More complex hands performing less complex but heavy-duty tasks will lead to faster wear and increased maintenance, also higher costs. Tesla is aiming for general-purpose applications.

I think we need a standard definition of ‘fully autonomous’

He gonna have back problems.

Miles ahead of Optimus radio doll

That’s great. Train it for underwater, or Lava. People can do that easily. For less than a billion dollars. EXPAND our capabilities NOT replace them

