Skip to content

VisionPilot v0.1.0-alpha — First VisionPilot Release

Latest
Latest

Choose a tag to compare

View all tags
@Julian1777 Julian1777 released this 26 Aug 15:43
· 1 commit to main since this release
v0.1.0-alpha
c21d863

VisionPilot v0.1.0 Release Notes

This is an experimental initial release of VisionPilot. It introduces baseline straight-line autonomous driving capability within the BeamNG.tech simulation environment alongside pre-trained perception models and core control features.

Release Highlights

  • Pre-trained Models: Includes model weights bundled directly in the visionpilot-v0.1.0-models.zip release archive.
  • Straight-Line Driving: Functional straight-line lane keeping, basic cruise control, and PID steering integration.
  • Perception & Assist Systems: Initial implementation of multi-lane detection, traffic sign/light detection and classification, Automatic Emergency Braking (AEB), and Cruise Control (CC).
  • Visualization: Integration with Foxglove Studio for real-time telemetry, camera feed, and LiDAR point cloud streaming.

Testing Environment

  • BeamNG.tech: Version 1.38
  • Python: Version 3.10
  • CUDA: Required for the perception models. CUDA acceleration is recommended for fast inference.
  • Foxglove Studio: Used for visualization and telemetry through its web interface.

Known Limitations

  • Curve Handling: Vehicle control currently only supports driving in a straight line or handling gentle path variations. Tight curves are not yet fully supported and will cause it to fail.

  • Low-Light Conditions: Perception reliability drops in tunnel or very low-light environments.

  • Single Camera Setup: Perception pipeline is currently limited to the front-facing camera.

  • Hardware Validation: This release has currently only been tested on the development machine used for the release. Testing across additional hardware configurations has not yet been performed.

What's Next

  • Curve Driving & Control: Tuning the PIDF steering controller to fix curve handling stability and implementing curve speed optimization.

  • High-Priority Features: Developing lane change logic (checking blind spots, signaling, executing) and introducing Birds-Eye View (BEV) spatial visualization.

  • Perception & Sensors: Adding road marking detection (arrows, crosswalks, stop lines), LiDAR object detection, speed estimation, and integrating GPS/IMU sensors.

  • Environmental Testing: Expanding evaluation across fog, low-light, and nighttime scenarios.

Setup Instructions

For detailed instructions on setting up and running VisionPilot, including installing dependencies, configuring the environment, downloading the pretrained models, and launching the system, please refer to the Usage section in the README.

The README also documents the models included in the models directory and their respective purposes.

Safety Notice: VisionPilot is an experimental research project intended for simulation and development purposes. It has not been validated for use with real vehicles.

Assets 4
Loading