Xingxin on Bug The good news is that my method finally works!! I am super happy!!! My effort on aligning to DROID dataset finally paid off!
But the bad news is… It only works for one specific task. I chose 4 other tasks for the demo and none of them worked out! What the heck… If that is the case, my paper might look very weak, as it only works for one specific contact-rich task…
However, this situation isn’t unfamiliar to me. My gut feeling tells me that the gripper issue is affecting my data distribution again. I modified a value, but I am not sure if it will work… I submitted the training job anyway… There are only 1.5 days left until the deadline… Am I still able to pull this off?
Finger crossed…🙏