Train, test, and validate perception systems with realism

Equipped with validated sensor models and procedural pipelines for synthetic 3D world creation, Spectral enables accurate sensor simulation for autonomous vehicle development.

Key features

Procedural generation of large-scale synthetic data for any target domain
Easy randomization and variation of environmental conditions (e.g., weather, lighting, road surfaces)
Deterministic simulation with precise and repeatable control of objects
Rich ground truth labels, annotations, and metrics
Diverse annotation types including 2D bounding boxes, cuboids, semantics, and more
Large library of validated sensor models (e.g., lidar, radar, camera)

Comprehensively test for edge cases and rare events

Create scenarios that are not possible or dangerous to test on the road. Vary and randomize environmental parameters such as weather and lighting for comprehensive testing.

Reduce data collection and labeling costs

Our pre-annotated synthetic data is significantly cheaper than on-road data that is collected and labeled manually. Use it for a variety of development use cases when immediate turnaround time is critical.

Improve perception performance

Use low-cost synthetic data for domains where real-world data collection is difficult and to increase coverage of long-tail situations.

Enable custom sensor models

Configure sensors easily and enable custom sensor models before hardware is available. Our sensor models are physics-based and model all external radiometric and internal signal processing processes and algorithms.