Flexibility and walking stability are greatly enhanced
Can accurately separate 4680 batteries
The rarest thing is that Tesla exposed the training of robots
It seems that training relies entirely on visual assistance and human demonstrations, which is the same logic as Tesla cars!
Tesla Engineers @milankovac More information was shared.
Their team built more robots, collected more AI data, and trained and deployed neural networks, allowing Optimus robots to start performing practical tasks. These tasks include picking up battery units from the conveyor belt and accurately placing them into trays.
Technical details:
The neural network used by Optimus runs completely end-to-end, using only the robot’s 2D camera video and onboard self-sensors to directly generate joint control sequences.
This neural network runs entirely on the robot’s embedded FSD computer and is powered by an onboard battery. The design allows a single neural network to perform multiple tasks after adding more diverse data during the training process.
performance characteristics
Although it is not perfect and a little slow at present, Optimus ‘success rate in performing tasks has gradually increased and failure rate has decreased.
In addition, the team is training Optimus to recover from failure and has observed spontaneous error correction behavior.
Optimus can already walk long distances in the office without falling and continues to test and improve on real workstations.
Further work:
The team is working to increase Optimus’s movement speed and handle more complex terrain without sacrificing its humanoid behavioral characteristics. At the same time, attention is also being paid to the repetitive capabilities of the entire robot population, training neural networks to handle dynamic calibrations and small inter-robot differences.
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Post link:https://x.com/_milankovac_/status/1787028644399132777
Video: