search

liufanfanlff / RoboUniview

39 stars 3 forks source link

Pretrained model without Flamingo Policy #1

Closed mbreuss closed 4 months ago

mbreuss commented 4 months ago

Hi,

thanks for sharing this cool work and code! The performance on ABC is very impressive. Did you ever test your model with another policy that is not a VLM? If so, what was the results?

liufanfanlff commented 4 months ago

Thank you for your attention and recognition of our work. We have tried combining it with 3D Diffuser Actor result as： | Method | CALVIN D-D Avg Len | | 3D Diffuser Actor（original） | 2.223 | | Uniview Diffuser Actor | 2.13 | We only made a simple attempt and did not conduct a detailed search for parameters and structures. If you conduct more experiments, may have better results.

mbreuss commented 4 months ago

Thanks for sharing!