Hi Rutav,
I have a question about the initial state used during evaluation. In your simulation experiments, do you reset the environment to the same initial state as the in-context demonstration, or do you randomly initialize the evaluation state (e.g., object poses, robot pose) independently of the demonstration?
If the evaluation uses random initialization, could you share the exact reset/randomization settings (which variables are randomized, their ranges/distributions)? If possible, it would also be very helpful if you could share the evaluation code/scripts you used.
Best,
Jian
Hi Rutav,
I have a question about the initial state used during evaluation. In your simulation experiments, do you reset the environment to the same initial state as the in-context demonstration, or do you randomly initialize the evaluation state (e.g., object poses, robot pose) independently of the demonstration?
If the evaluation uses random initialization, could you share the exact reset/randomization settings (which variables are randomized, their ranges/distributions)? If possible, it would also be very helpful if you could share the evaluation code/scripts you used.
Best,
Jian