ONRAP: Occupancy-driven Noise-Resilient Autonomous Path Planning

TL;DR

A practical, implementation-friendly planner that operates on occupancy grids and optionally incorporates occupancy-flow predictions to generate ego-centric, kinematically feasible paths that safely navigate through static and dynamic obstacles.

Abstract

Dynamic path planning must remain reliable in the presence of sensing noise, uncertain localization, and incomplete semantic perception. We propose a practical, implementation-friendly planner that operates on occupancy grids and optionally incorporates occupancy-flow predictions to generate ego-centric, kinematically feasible paths that safely navigate through static and dynamic obstacles. The core is a nonlinear program in the spatial domain built on a modified bicycle model with explicit feasibility and collision-avoidance penalties. The formulation naturally handles unknown obstacle classes and heterogeneous agent motion by operating purely in occupancy space. The pipeline runs in real-time (faster than 10 Hz on average), requires minimal tuning, and interfaces cleanly with standard control stacks. We validate our approach in simulation with severe localization and perception noises, and on an F1TENTH platform, demonstrating smooth and safe maneuvering through narrow passages and rough routes. The approach provides a robust foundation for noise-resilient, prediction-aware planning, eliminating the need for handcrafted heuristics. The project website can be accessed at https://honda-research-institute.github.io/onrap/

Figures (9)

• Figure 1: Motivation. The ego vehicle (green) must deviate from the lane center to avoid parked vehicles (gray) along the roadside while remaining aware of oncoming traffic (red). Blurred vehicles indicate perception uncertainty. The planner processes the occupancy grid, with occupied cells highlighted in red, and generates the blue trajectory, enabling safe navigation around obstacles and dynamic traffic, in the presence of real-world uncertainties.
• Figure 2: Pipeline. Raw sensory input data is processed by the BEV generation modules to produce BEV data for the occupancy grid generator, eliminating the need for intermediary semantic segmentation and object recognition modules. The resulting occupancy grid is then fed into our ONRAP module, which computes a kinematically feasible path for the downstream speed planning module.
• Figure 3: Ego-centric occupancy grid. The ego-centric occupancy grid is represented as a fixed-size matrix centered on the ego vehicle. Arbitrary occupancies from the global map are projected into this local frame, preserving spatial relationships while ensuring constant computational complexity regardless of obstacle count or shape.
• Figure 4: Pareto analysis of the risk function $\exp{\left(-\frac{(y_k-y_{\mathcal{G}})^2}{2(\sigma\cdot\tau)^2}\right)}$ for different $(\sigma,\tau)$ pairs. Here, $\sigma$ represents the desired safety distance, while $\tau$ controls the sharpness of the risk profile (smaller $\tau$ yields a steeper decay).
• Figure 5: Path planning results in the validation scenario. The ego vehicle (green star) navigates a sinusoidal route (yellow) without access to ground-truth values. The magnified regions highlight two primary challenges, among others: (i) obstacles located at regions of high curvature, and (ii) a narrow passage between occupied cells with no direct free space along the route. The cyan line represents the simulated trajectory generated by the planner.

Theorems & Definitions (7)

• Remark 1
• Remark 2
• Remark 3
• Remark 4
• Remark 5
• proof
• Remark 6