Issue with Multiple render_offscreen Calls Using Different Dimensions in MuJoCo

[cherylwang20]

Hi,

I am a student currently utilizing MuJoCo for reinforcement learning (RL) training and encountering a problem with rendering from different camera perspectives.

Problem Description:
I need to use the render_offscreen method for two purposes:

1. To capture high-quality video output for visualization and presentation.
2. To provide input for a CNN-based policy as part of the training process.

Here is a code snippet demonstrating how I am currently handling these tasks:

# Outputting to video rendering
rgb_out = self.sim.renderer.render_offscreen(width=800, height=800, camera_id='ft_cam')
self.sim.renderer.close()

# For CNN training purposes with dimensions 224x224
rgb, depth = copy.deepcopy(
    self.sim.renderer.render_offscreen(width=width, height=height, camera_id=camera, depth=True)
)

Issue:
The problem arises when attempting to use render_offscreen with different dimensions sequentially. Whichever render_offscreen call is executed first sets the dimensions for subsequent calls within the same simulation instance. This leads to the second render_offscreen call not initializing a new screen, but instead using the previous dimensions specified.

Request for Assistance:
Could someone please provide guidance on how to correctly reinitialize or reset the renderer so that it correctly handles multiple calls to render_offscreen with different dimensions? Any advice or examples of how to manage this would be greatly appreciated.

Thank you for your assistance!