Rasterizer#

The Rasterizer provides fast GPU-accelerated rendering using OpenGL. It’s the default renderer for real-time visualization and is suitable for training reinforcement learning agents.

Overview#

The Rasterizer offers:

High performance: 1000+ FPS for simple scenes
GPU acceleration: Leverages OpenGL for rendering
Multiple outputs: RGB, depth, segmentation, normals
Multi-environment support: Efficient batched rendering

Quick Start#

import genesis as gs

gs.init()
scene = gs.Scene()

# Add entities
scene.add_entity(gs.morphs.Plane())
robot = scene.add_entity(gs.morphs.URDF(file="path/to/robot.urdf"))

# Add rasterizer camera (default)
cam = scene.add_camera(
    res=(640, 480),
    pos=(3, 0, 2),
    lookat=(0, 0, 0.5),
    fov=40,
)

scene.build()

# Render images
for i in range(100):
    scene.step()

    rgb, _, _, _ = cam.render(rgb=True)
    _, depth, _, _ = cam.render(depth=True)
    _, _, segmentation, _ = cam.render(segmentation=True)

Configuration#

The Rasterizer is configured through RasterizerOptions:

rasterizer_options = gs.options.renderers.Rasterizer(
    env_separate_rigid=True,   # Separate rendering per environment
)

Output Types#

Output	Type	Description
`rgb`	`np.ndarray` (H, W, 3)	RGB color image
`depth`	`np.ndarray` (H, W)	Depth values in meters
`segmentation`	`np.ndarray` (H, W)	Entity/link segmentation IDs
`normal`	`np.ndarray` (H, W, 3)	Surface normals

API Reference#

class genesis.vis.rasterizer.Rasterizer(viewer, context)[source]#

Bases: RBC

build()[source]#

add_camera(camera)[source]#

update_camera(camera)[source]#

remove_camera(camera)[source]#

render_camera(camera, rgb=True, depth=False, segmentation=False, normal=False)[source]#

update_scene(force_render: bool = False)[source]#

destroy()[source]#

property viewer#

property offscreen#