A modular autonomous driving platform for developing and testing AV components on CARLA simulator and real-world vehicles.
Pylot is an open-source, modular platform for developing and testing autonomous vehicle components like perception, prediction, planning, and control. It enables researchers and engineers to experiment with algorithms in the CARLA simulator and on real-world vehicles, providing tools for data collection, component isolation, and system integration. The platform addresses the need for a flexible, reproducible environment to explore latency-accuracy tradeoffs in autonomous systems.
Autonomous vehicle researchers, robotics engineers, and developers working on self-driving car algorithms who need a modular platform for simulation, testing, and real-world deployment.
Pylot offers a unique combination of modularity, multi-environment support (CARLA and real vehicles), and configurability, allowing users to swap components, test in isolation, and benchmark against the CARLA Leaderboard. Its open-source nature and integration with the ERDOS framework enable rapid prototyping and exploration of autonomous driving stacks.
Modular autonomous driving platform running on the CARLA simulator and real-world vehicles.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Components like obstacle detection and planning are implemented as independent ERDOS operators, allowing isolated testing or full integration, which accelerates debugging and algorithm development.
Seamlessly runs on the CARLA simulator with data collection scripts for RGB images, point clouds, and more, enabling safe, scalable testing and benchmarking on the CARLA Leaderboard.
Offers multiple planners (e.g., Frenet optimal trajectory, RRT*) and controllers (e.g., PID, MPC), allowing users to experiment with different driving strategies for various scenarios.
Can be deployed on physical vehicles, facilitating validation from simulation to reality, as mentioned in the platform's multi-environment support.
Requires nvidia-docker, specific CARLA version installation, and multiple steps for visualization, making the onboarding process time-consuming and error-prone.
Only supports a few detection models (e.g., ssd-mobilenet-fpn-640), and users must rely on 'perfect' versions or custom training for advanced accuracy, as admitted in the README.
Heavily tied to CARLA simulator; adapting to other platforms or real-world setups requires significant modifications, limiting flexibility for broader use cases.