Embodied AI Glossary中文

Path Tracing

路径追踪Advanced

A rendering method that simulates light bouncing many times through a scene using randomly sampled rays, producing physically realistic images.

Path tracing was introduced by Jim Kajiya in 1986, in the same paper that gave the rendering equation (the integral equation describing how light propagates through a scene) and approximated it with Monte Carlo integration (averaging over random samples). It works by casting a ray from the camera through each pixel, and whenever it hits a surface, randomly choosing a new direction to keep bouncing based on the material, accumulating lighting along the way — which naturally produces effects like indirect illumination, soft shadows, and caustics. The cost is that images are noisy when too few samples are used, requiring either very heavy sampling or a neural denoiser to clean up. It belongs to the ray-tracing family and is much slower than rasterization, traditionally reserved for offline rendering such as film. In robot simulation, the Omniverse RTX renderer used by Isaac Sim offers a path-tracing mode that produces more photorealistic synthetic images, at the cost of being slower than its real-time mode.

ExampleOmniverse's RTX Interactive (Path Tracing) mode defaults to sampling 1 ray per pixel per frame, accumulating up to 512 samples, with the OptiX denoiser enabled by default.

Also called
Path-Traced Rendering, Monte Carlo Ray Tracing
Related
Ray Tracing · Rasterization · Physically Based Rendering · Photorealistic Rendering · Rendering · NVIDIA Isaac Sim
Sources
Wikipedia: Path tracing
Omniverse Docs: RTX Interactive (Path Tracing) Mode
Physically Based Rendering: From Theory to Implementation, 4th ed. (online)
As of
2026-09

See it in the full glossary →