Frenet Corridor Planner: An Optimal Local Path Planning Framework for Autonomous Driving

TL;DR

The Frenet Corridor Planner (FCP) addresses the need for real-time, reliable local path planning for autonomous driving within a path–speed decomposition framework. It constructs a drivable corridor by representing obstacles as safety-augmented bounding boxes and pedestrian convex hulls in Frenet space, then optimizes a path using a space-domain bicycle kinematics model with curvature-based actuation limits, and finally hands the path to a speed planner. Key contributions include a modular pipeline (DP, DG, BG, PO) with a low-complexity boundary generation, a convex-structured optimization with dynamic obstacle risk and perception-noise handling, and extensive validation in CARLA, Monte Carlo studies, and scaled hardware demonstrations. The results demonstrate improved runtime, smoother trajectories, and enhanced safety margins, supporting real-time deployment in urban driving scenarios.

Abstract

Motivated by the requirements for effectiveness and efficiency, path-speed decomposition-based trajectory planning methods have widely been adopted for autonomous driving applications. While a global route can be pre-computed offline, real-time generation of adaptive local paths remains crucial. Therefore, we present the Frenet Corridor Planner (FCP), an optimization-based local path planning strategy for autonomous driving that ensures smooth and safe navigation around obstacles. Modeling the vehicles as safety-augmented bounding boxes and pedestrians as convex hulls in the Frenet space, our approach defines a drivable corridor by determining the appropriate deviation side for static obstacles. Thereafter, a modified space-domain bicycle kinematics model enables path optimization for smoothness, boundary clearance, and dynamic obstacle risk minimization. The optimized path is then passed to a speed planner to generate the final trajectory. We validate FCP through extensive simulations and real-world hardware experiments, demonstrating its efficiency and effectiveness.

Figures (10)

• Figure 1: Motivational scenario. The ego vehicle (in green) must deviate from the lane center to avoid a collision with the parked cars on the roadside while being cognizant of the oncoming traffic. Without a local path planner, the ego vehicle may remain stuck, waiting indefinitely for the parked cars to move before proceeding along its pre-determined global route.
• Figure 2: Trajectory planning pipeline. The data flow between the various building blocks of FCP is illustrated on the left, while the output visualization from each module is shown on the right.
• Figure 3: Decision tree for boundary classification of each obstacle. The decision tree evaluates the lower and upper gaps within the drivable space. If both gaps are available, two approaches can be used: selecting the preferred gap based on cost evaluation or treating the obstacle as a risk in PO.
• Figure 4: Boundary Generation. With the pedestrians shown as blue dots, the augmented vehicle boundary depicted by green dots, and the convex hulls of pedestrian clusters given by the blue lines, Algorithm \ref{['alg:boundary_generation']} generates the lower and upper bounds, shown by the red and blue lines, respectively.
• Figure 5: Numerical validation for $\mathbf{\frac{l_r}{l_f+l_r}|\delta_k|}$ under-approximating $\mathbf{|\beta_k|}$. The linear plot $\frac{l_r}{l_f+l_r}\delta_k$ stays below the $\beta_k$ curve for $d_k \in [0,\frac{\pi}{2})$ and above $\beta_k$ for $d_k \in (-\frac{\pi}{2},0]$ showing $|\beta_k| \geq \frac{l_r}{l_f+l_r}|\delta_k|~\forall \delta_k \in (-\frac{\pi}{2},\frac{\pi}{2})$.

Theorems & Definitions (2)

• Remark 1
• proof