EagleVision

Unified LiDAR Perception for High-Speed Autonomous Racing

EagleVision is a LiDAR-based perception benchmark spanning real-world competition data from the Abu Dhabi Autonomous Racing League (A2RL) and the Indy Autonomous Challenge (IAC), together with high-fidelity simulator ground-truth. All domains are standardized into a common coordinate and annotation convention for consistent cross-domain evaluation.

Get the Dataset View Dataset Details
Competition Vehicle
Autonomous racing vehicle

What’s inside

3 Domains
IAC, A2RL Simulator, A2RL Real-World
  • Cross-domain evaluation with a unified convention
  • Real racing conditions from competition recording
  • Simulator ground-truth for scalable training
Tasks
Detection + Trajectory Prediction
  • 3D object detection (single class: Car)
  • Trajectory prediction from frame-level pose time-series
  • Standard splits + baseline-friendly formats
Unified Format
PSR JSON schema (Position-Scale-Rotation)
  • Center (x,y,z), size (l,w,h), yaw
  • Ego-vehicle LiDAR frame
  • No motion compensation (consistent across real & sim)

Dataset at a glance

Domain LiDAR Hz Annotated frames Avg objects/frame Points/scan (approx.)

Numbers are configurable in data/dataset_stats.json.