FRENETIX: A High-Performance and Modular Motion Planning Framework for Autonomous Driving

TL;DR

FRENETIX introduces a modular, high‑performance sampling‑based trajectory planner for autonomous driving that operates in the Frenet coordinate system to efficiently handle dynamic urban and highway environments. The framework combines environment preprocessing from lanelet/CommonRoad data, an iterative trajectory planning cycle with polynomial trajectory generation, multi‑objective cost evaluation, and rigorous validity checks, all implemented in Python with performance‑critical paths in C++. It achieves real‑time capability (e.g., 8 ms for 800 trajectories) and demonstrates an 88% success rate across 1750 complex scenarios, with further validation on a real research vehicle via ROS2 and a CommonRoad interface. By open‑sourcing FRENETIX and its modular extensions, the work provides a versatile baseline for benchmarking, experimentation, and integration with perception and control pipelines, enabling future exploration of behavior planning and reinforcement‑learning based weighting in autonomous driving.

Abstract

Our research introduces a modular motion planning framework for autonomous vehicles using a sampling-based trajectory planning algorithm. This approach effectively tackles the challenges of solution space construction and optimization in path planning. The algorithm is applicable to both real vehicles and simulations, offering a robust solution for complex autonomous navigation. Our method employs a multi-objective optimization strategy for efficient navigation in static and highly dynamic environments, focusing on optimizing trajectory comfort, safety, and path precision. The algorithm is used to analyze the algorithm performance and success rate in 1750 virtual complex urban and highway scenarios. Our results demonstrate fast calculation times (8ms for 800 trajectories), a high success rate in complex scenarios (88%), and easy adaptability with different modules presented. The most noticeable difference exhibited was the fast trajectory sampling, feasibility check, and cost evaluation step across various trajectory counts. We demonstrate the integration and execution of the framework on real vehicles by evaluating deviations from the controller using a test track. This evaluation highlights the algorithm's robustness and reliability, ensuring it meets the stringent requirements of real-world autonomous driving scenarios. The code and the additional modules used in this research are publicly available as open-source software and can be accessed at the following link: https://github.com/TUM-AVS/Frenetix-Motion-Planner.

Figures (12)

• Figure 1: Visualization of a trajectory planner for AVs, depicting potential trajectories while selecting an optimal trajectory for safe navigation in a dynamic and complex urban environment.
• Figure 2: Simplified trajectory planning procedure.
• Figure 3: Visualization of the sequential progression of the motion planning algorithm over time. The black trajectory shows the selected optimal trajectory. The gray trajectories are kinematically infeasible. The feasible trajectories are highlighted using a color scale corresponding to the cost, where green indicates a low cost and red indicates a high cost.
• Figure 4: Frenet coordinate system for trajectory generation, adapted from werling.2012.
• Figure 5: Sampling process in different coordinate systems, as per werling.2010: subfigure (a) illustrates trajectories in curvilinear coordinates, capturing the primary direction of movement within a curved reference frame. Subfigure (b) translates these trajectories into cartesian coordinates. The ego vehicle is blue ( ), indicating its position following the sampled trajectories along the reference path ( ).