ONRAP: Occupancy-driven Noise-Resilient Autonomous Path Planning

Faizan M. Tariq, Avinash Singh, Vipul Ramtekkar, Jovin D'sa, David Isele, Yosuke Sakamoto, Sangjae Bae
Honda Research Institute, USA
2026 IEEE Intelligent Vehicles Symposium (IV 2026)
Paper Code*

*The ONRAP software is not publicly available; however, its core architecture is built upon the FCP framework.

📘 Overview

Autonomous systems often operate under imperfect sensing: detections may be noisy, object classes may be wrong or unavailable, and localization can drift. ONRAP addresses these challenges by planning directly in occupancy space rather than relying on object-centric scene representations.

Given an ego-centric occupancy grid and a local goal, ONRAP generates a smooth reference path and optimizes a feasible trajectory that balances reference tracking, collision risk, control effort, and curvature. Static and dynamic obstacles are represented uniformly through occupancy, while optional occupancy-flow estimates allow the planner to account for predicted motion.

ONRAP: noise-resilient, occupancy-driven path planning across simulation and hardware

🧩 Key Features

  • Occupancy-first planning
    Uses occupancy grids as the primary safety signal, enabling class-agnostic reasoning over obstacles without relying on object labels or semantic classes.
  • Noise-resilient formulation
    Handles noisy reference paths, localization drift, and uncertain perception inputs, remaining robust when sensing or semantic understanding is imperfect.
  • Kinematically feasible paths
    Formulates local path generation as a spatial-domain nonlinear program based on bicycle kinematics, ensuring smooth and drivable outputs.
  • Prediction-aware behavior
    Can incorporate occupancy-flow estimates to anticipate dynamic obstacles and plan proactively.
  • Real-time deployment
    Demonstrated at over 10 Hz on average, including experiments on an F1TENTH platform with limited onboard compute.
  • Minimal tuning
    Transfers across simulation and hardware experiments with only vehicle-specific parameters adjusted.

📊 Evaluation

ONRAP is evaluated in noisy simulation environments and on a real F1TENTH vehicle platform. The experiments include cluttered maps, narrow passages, high-curvature routes, noisy occupancy inputs, noisy reference paths, and moving obstacles.

Compared with A* and RRT* baselines in randomized simulations, ONRAP achieves higher success rates, better obstacle clearance, faster runtime than RRT*, and substantially smoother trajectories.

📚 Citation

If you find this work useful, please cite:

@misc{tariq2026onrap,
  title         = {ONRAP: Occupancy-driven Noise-Resilient Autonomous Path Planning},
  author        = {Tariq, Faizan M. and Singh, Avinash and Ramtekkar, Vipul and D'sa, Jovin and Isele, David and Sakamoto, Yosuke and Bae, Sangjae},
  year          = {2026},
  eprint        = {2602.13577},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  doi           = {10.48550/arXiv.2602.13577},
  note          = {Presented at the 2026 IEEE Intelligent Vehicles Symposium}
}