[Question] Using imitation learning for training

[longann890621]

Question

Hello, I have the following two questions regarding training in imitation learning：

1. I used the following commands to train my Franka task (stack_ik_rel_visuomotor_env_cfg.py):

./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py
--task Isaac-Stack-Cube-Franka-IK-Rel-Visuomotor-v0
--algo bc
--dataset ./datasets/franka_vision_final3.hdf5

If I want to change the training project to include the existing "joint_pose" for this task, I simply need to change bc_rnn_image_200.json to：
"observation": {
"modalities": {
"obs": {
"low_dim": [
"eef_pos",
"eef_quat",
"gripper_pos"
"joint_pos"
],
"rgb": [
"table_cam",
"wrist_cam"
],

What other parts of the code do I need to modify to successfully complete the training?Or is this sufficient?

---

2.If I have a brand new task that I set myself today, I will use the following command：

./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py
--task My-Franka-mission
--algo bc
--dataset ./datasets/My_franka.hdf5

I can create a completely new gym.register in init.py, then point it to a completely new JSON object, and rewrite it as follows：
"observation": {
"modalities": {
"obs": {
"low_dim": [
"AAA_pos",
"AAA_quat",
"gripperAAA_pos"
],
"rgb": [
"tableAAA_cam",
"wristAAA_cam"
],

Will this approach allow me to complete custom task training?