First, congratulations on the excellent work with RDT-1B! 🥳 Your approach to diffusion-based bimanual manipulation and the introduction of a unified action space are both impressive and inspiring.
I am interested in adapting your method for use on a different robotic setup, specifically a UR5 robotic arm. I have a few questions regarding this adaptation:
Compatibility: Given the unified action space you've introduced, would it be feasible to apply RDT-1B on a UR5 arm setup? Are there specific adjustments or considerations you would recommend?
Fine-Tuning Requirements: Assuming additional fine-tuning is needed for optimal performance on UR5, could you provide guidance on the approximate data volume and computational resources (e.g., GPU requirements) necessary for fine-tuning?
Any insights or recommendations would be greatly appreciated. Thank you again for your work on this exciting project! 🙏
Compatibility: Yes, it is compatible. No adjustment is needed. You can control a UR5 with either a joint-based or EEF-based controller.
Fine-Tuning Requirements: At least 24GB VRAM.
Hi,
First, congratulations on the excellent work with RDT-1B! 🥳 Your approach to diffusion-based bimanual manipulation and the introduction of a unified action space are both impressive and inspiring.
I am interested in adapting your method for use on a different robotic setup, specifically a UR5 robotic arm. I have a few questions regarding this adaptation:
Compatibility: Given the unified action space you've introduced, would it be feasible to apply RDT-1B on a UR5 arm setup? Are there specific adjustments or considerations you would recommend?
Fine-Tuning Requirements: Assuming additional fine-tuning is needed for optimal performance on UR5, could you provide guidance on the approximate data volume and computational resources (e.g., GPU requirements) necessary for fine-tuning?
Any insights or recommendations would be greatly appreciated. Thank you again for your work on this exciting project! 🙏