[Feature]: Parallel Eye-in-Hand Camera Simulation #1232
Description
What feature or enhancement are you proposing?
I would like a feature where I can attach a camera to a robot link prior to building the scene. Then when I build, each robot has its own camera attached that follows their respective robot. Then I can render each camera and fetch the data in batched commands.
This would look similar to the already implemented cam.attach function, but being placed before building the scene.
Motivation
Eye-in-hand camera are one of the most common setups for robotic manipulators. Currently there is no way to efficiently simulate an eye-in-hand camera per environment.
Potential Benefit
This will open up genesis to many researchers that are attempting to implement reinforcement learning with image-based robotics.
What is the expected outcome of the implementation work?
- Attach cameras prior to building a scene that get copied per environment
- Cameras automatically move with their respective attached link in each environment
- Can get camera data in batched returns i.e [camera data, n_envs]
Additional information
No response