Step is a versatile synthetic data generation API that allows users to fine-tune the environment, lighting, sensor configuration, and agent behavior of their synthetic sensor data. With Step, users can achieve two primary use cases:

1. Closed-Loop Perception Testing:

Step can be integrated with simulation stacks to generate synthetic sensor data on-demand, enabling users to bring perception testing to their simulation workflows and run large-scale simulations to test different scenarios. This allows users to customize every detail of the scenario and test their perception systems with precision.

Step also provides client libraries in Python and C++ to integrate with users' simulators, as well as a cloud-hosted render pipeline that returns synthetic sensor data based on transmitted state information.

2. Surgical Modification and Generation of Synthetic Data:

Step grants users deep control over sensor data on a frame-by-frame basis. Users can insert or modify desired sensor data, including agents, behaviors, weather, lighting, and sensor configurations. These modifications can then be scaled across other frames and scenes and generated via Batch or Step. Examples of use cases include:

  • Adding a group of pedestrians jaywalking at an intersection 5 meters away from the ego vehicle and applying the same modification to a thousand other frames.
  • Inserting emergency vehicles with sirens on in every night scene.
  • Adding three cyclists (one on each side and at the front of the ego vehicle) and changing the weather to rainy for every day scene.

There are two main components for the Step API.

  • The Step Management API is a REST Endpoint used to create, list and delete Step server instances.

  • The Application API (or Step API) allows you communicate with a single Step server instance. It allows you to:

    • Construct agent objects and transmit their state to the StepServer.
    • Programmatically create camera and LiDAR sensors directly from code.
    • Access Parallel Domain's library of carefully curated and calibrated content.