@Tesla_Optimus team behind optimus (figuratively and literally)!
Ashish
77 posts
- Sim-to-real RL. zero-shot. Everything is running onboard and realtime!
- Learned with RL in sim, deployed zero shot to real We are planning to extend our RL sim training to manipulation in the coming months (with 22 dof hands). Initial results will produce dexterous robot specialists. My team is also building a generalist robot policy byAccurate actuators accelerate automation
00:00 - Here are some more! With music. All of these use the same pipeline: RL training in sim followed by sim2real (zero-shot). Our pipeline is relatively robust and requires minimal tuning per dance. This allows exponential scaling to new behaviors!
- Optimus on charging break during 10/10! What people missed about the ~20 human-assisted @Tesla_Optimus is Tesla’s intent & ability to scale up data collection (~20x more than 2022 AI Day). Language, images, audio are blessed with exact data, robotics will need to build it out
- We are making the @Tesla_Optimus walk increasingly more robust and ready for all terrains. Achieved by replacing a chunk of c++ code with nets!
- Just graduated from @UCBerkeley, now at @Tesla! In my PhD, I showed the effectiveness of pure learning for: -Locomotion (tinyurl.com/3k6mtrac) -Dexterous manipulation (tinyurl.com/ywtjtwwh) -Universal drone flight (tinyurl.com/ye28tkjz) Next up: humanoids! (@Teslasbot)
00:00 - Autonomy update from Optimus! Autonomous navigation, self charge, snack handout, stairs — none of it is tele-operated in this video.Navigating by myself
00:00 - All tasks are from the same neural net that understands text instructions! Our technical breakthrough is in directly learning from first person videos of humans doing tasks! This allows us scale to new skills much faster vs using robot data only.
- Easily the most beautiful and elegant hardware I have seen. By a margin. Check out our new humanoid build at @Tesla and what its capable of!
- A sneak peek of what we have been up to at @Tesla with the humanoid robot -- @Tesla_Optimus ! A robust controller is critical to real-world deployment -- we get it as a natural consequence of using end-to-end learning with large scale data!Optimus can now sort objects autonomously 🤖 Its neural network is trained fully end-to-end: video in, controls out. Come join to help develop Optimus (& improve its yoga routine 🧘) → tesla.com/AI
00:00 - It is indeed quite magical seeing Optimus randomly drop by outside your meeting room! Also routinely walks amongst us, everyday, all day.I forgot to mention that Optimus was doing this while “blindfolded”. We haven’t yet made vision part of the control loop for uneven terrain. Optimus robots are regularly cruising around our workplace in Palo Alto where vision is used for static & dynamic object avoidance.
