autowarefoundation/autoware_vision_pilot
Free and fully open-source L2 ADAS stack powered by End-to-End AI technology
What it solves
Vision Pilot provides a productionizable and safety-certifiable implementation of Level 2 Advanced Driver Assistance Systems (ADAS). It enables in-lane autonomous driving features without the need for high-definition 3D maps, allowing vehicles to follow roads in real-time using only a single front-facing monocular camera.
How it works
The system utilizes a hybrid end-to-end AI architecture. It processes data in parallel using two distinct approaches: perception AI models for safety and end-to-end AI models for performance. The core logic relies on three specialized models: AutoSpeed for object detection, AutoSteer for waypoint detection, and AutoDrive for road curvature estimation and object presence detection.
Who it’s for
This codebase is designed for automotive OEMs and Tier-1 suppliers for use in series production passenger cars, as well as transportation and logistics providers using buses and trucks.
Highlights
- Supports features like Adaptive Cruise Control (ACC), Autonomous Emergency Braking (AEB), and Lane Keep Assist (LKAS).
- Operates in a 'apless' mode using only a single monocular camera.
- Includes open-source AI model weights to democratize access to the technology.
- Compatible with the CARLA simulator for closed-loop testing.
- Available under the permissive Apache 2.0 license.