How to adjust the camera angle in the output video to show more of the robotic arm?

[wugh]

Hi, I want to camera angle in the output video to show more of the robotic arm, but dont want to change the input of inference model, how to do it?

[xuanlinli17]

For visual matching evaluation, unfortunately this is not possible to do (since we overlay the real background into simulated observations)

For variant aggregation evaluation, you can render images from a 3rd camera view. You can adjust the "human" camera config in the code.

[wugh]

For visual matching evaluation, unfortunately this is not possible to do (since we overlay the real background into simulated observations)

For variant aggregation evaluation, you can render images from a 3rd camera view. You can adjust the "human" camera config in the code.

thank you for your reply xuanlin,

I think what you mean is script like scripts/octo_pick_coke_can_variant_agg.sh ? I didn't find option to adjust the camera config, so this camera config is hard code in some file?

[xuanlinli17]

Yes.

You can add cameras to the robot config file like https://github.com/simpler-env/ManiSkill2_real2sim/blob/cd45dd27dc6bb26d048cb6570cdab4e3f935cc37/mani_skill2_real2sim/agents/configs/google_robot/defaults.py

[wugh]

Yes.

You can add cameras to the robot config file like https://github.com/simpler-env/ManiSkill2_real2sim/blob/cd45dd27dc6bb26d048cb6570cdab4e3f935cc37/mani_skill2_real2sim/agents/configs/google_robot/defaults.py

ok, I will try this. thank you