Open quanfeifan opened 6 months ago
An alternative code is now avail as Octo 🐙 https://github.com/octo-models/octo
Hi! The Octo models are not RT-1-Jax. We are hoping to release RT-1-Jax soon in the new year. Please stay tuned!
Best, Pannag
On Tue, Dec 19, 2023 at 8:59 PM 'youliang' via Open X-Embodiment Collaboration @.***> wrote:
The code is now avail as Octo 🐙 https://github.com/octo-models/octo
— Reply to this email directly, view it on GitHub https://github.com/google-deepmind/open_x_embodiment/issues/27#issuecomment-1863854142, or unsubscribe https://github.com/notifications/unsubscribe-auth/BDPTRVND35S3XWPRPBGRZU3YKJWBNAVCNFSM6AAAAABAKLKGL6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNRTHA2TIMJUGI . You are receiving this because you are subscribed to this thread.Message ID: @.***>
-- You received this message because you are subscribed to the Google Groups "Open X-Embodiment Collaboration" group. To unsubscribe from this group and stop receiving emails from it, send an email to @.*** To view this discussion on the web visit https://groups.google.com/d/msgid/open-x-embodiment/google-deepmind/open_x_embodiment/issues/27/1863854142%40github.com https://groups.google.com/d/msgid/open-x-embodiment/google-deepmind/open_x_embodiment/issues/27/1863854142%40github.com?utm_medium=email&utm_source=footer .
Just to second Pannag's point here: Octo is not the same as RT-1-X -- the models use different architectures, training mixes etc. The RT-1-X code should be out soon hopefully! In the meantime you could take a look at the Octo models if you want to start some finetuning on your robot.
Thanks to others for the further clarification :+1:
Hi,
thank you for open-sourcing RT-1 and the checkpoints. I'm very interested in this work and can't wait to fine-tune it in my real robot.
Could you tell me when the Jax version of the RT-1-X model will be open sourced?
Thanks!