Drake
Rendering Engines

Detailed Description

Rendering engines are used to enable simulation of the perception stack.

Overview

Perception simulation is supported via the RenderEngine API. It provides the interface through which the geometries in the simulation can be rendered to three types of images: color, depth, and label. (For more details about the API, refer to the RenderEngine documentation.)

Drake includes two implementations of that API:

These implementations differ in many ways largely differ in performance and flexibility potential. RenderEngineVtk is slower but has the possibility of leveraging the full VTK API to provide post hoc customizations. RenderEngineGl is faster, but gains that performance boost by implementing a barebones rendering pipeline.

Picking the right renderer for your simulated sensors will be based on considering those differences and picking the trade-off that best suits your application.

The current API is simple and precludes sophisticated definitions of lights and materials. As the API matures, these additional features will be added.

Performance

The performance of both of the current implementations will strongly depend on the quality of a system's GPU and its driver. A more powerful graphics card with an up-to-date driver will likely produce images more quickly.

It is impossible to state what the a RenderEngine's performance is in absolute terms. Instead, Drake includes a simple benchmark which can be run locally to assess the actual performance the end user will experience on a fixed system.

Modules

 Render Benchmarks
 Benchmarks for assessing rendering performance and comparing implementations.