Is it possible to train on the official dataset using imitation learning and then directly validate in real-world scenarios, even if the robotic arm is different from Franka? Would it be necessary to collect a new dataset?
If you are not using Franka, it would be interesting to see if fine-tuning on your new embodiment data would work. In general, it's recommended you collect a new dataset, and you can contribute to FMB
Is it possible to train on the official dataset using imitation learning and then directly validate in real-world scenarios, even if the robotic arm is different from Franka? Would it be necessary to collect a new dataset?