@ClaraChengGo Interesting take, I’d disagree with some points. Firstly I think model based and hardware based robotics companies should be treated differently. Hardware is similar to infra with slow cycle times and wouldn’t be easy to switch. Model companies have little first mover advantage
@gupta_abhinav_ What is the Omnibodied intelligence at Skild?
1. Astra style model with code as policy?
2. Pi API style base model + Fine tuning as a service?
3. One model with embodiment defining prompts
4. World action models?
5. Something else?
Do navigation and manipulation share gains?
From assembly to data collection to training and inference — all within 24 hours, across two different robots.
Guess the weekend was fun.
Honestly surprised by how fast Physical Intelligence’s cloud inference was.
We taught the butter robot to pass more than butter!
@GauravSethia8, Siddharth Ghodasara and I fine tuned @physical_int pi 0.7 model across two embodiments (so101 + butter robot), it’s just one policy doing both!
Had a great time at @ActorLabs act-athon