PyBatchRender

Fast batch 3D rendering for TorchRL environments using Panda3D. Try the CartPole demo in Colab to get started.

Installation

pip install git+https://github.com/dolphin-in-a-coma/pybatchrender.git

From source:

pip install -e .

Optional extras:

pip install "pybatchrender[images]"  # For saving frames
pip install "pybatchrender[cuda]"    # For CUDA interop (Linux/Windows with NVIDIA)

Quickstart

With pre-made environments:

import pybatchrender as pbr

env = pbr.envs.make("CartPole-v0", num_scenes=256)
td = env.reset()

for step in range(100):
    td["action"] = env.action_spec.rand()
    td = env.step(td)
    td = td["next"]

Render simple scene:

from pybatchrender import PBRConfig, PBRRenderer

cfg = PBRConfig(num_scenes=4, tile_resolution=(64, 64), offscreen=True)
renderer = PBRRenderer(cfg)

renderer.add_camera()
renderer.add_light()
renderer.add_node('models/box', instances_per_scene=1, shared_across_scenes=True)
renderer.setup_environment()

img_batch = renderer.step(return_pixels=True)
print(img_batch.shape)  # torch.Size([4, 3, 64, 64])

License

MIT