The two models you open-sourced in your project, octo-based and octo-small, were both trained on robotic arm data. Would it be better to fine-tune these two models on other configurations of robots, such as wheeled or legged robots, rather than “learning from scratch”? Thanks!
The two models you open-sourced in your project, octo-based and octo-small, were both trained on robotic arm data. Would it be better to fine-tune these two models on other configurations of robots, such as wheeled or legged robots, rather than “learning from scratch”? Thanks!