Safe Occlusion Aware Planning

This repository contains the experiment code for our RSS 2021 paper, Safe Occlusion-aware Autonomous Driving via Game-Theoretic Active Perception.

Usage and Limitations

In order to reproduce our results, you need to install a modified octomap-python and opendrive2lanelet packages provided in the ThridParty folder.

In addition, you will need Carla to run the simulation. We tested our code using Carla 0.9.11 with Python3.6 on Ubuntu 18.04.

You can run test _xxx.py file to see different examples.

This code is intended as a proof-of-concept demo of our proposed framework, and as such it leaves significant room for improvement on several fronts:

1. The occluded regions are currently deteced using a brute-force method.
2. The safe sets are calculated in closed-form with simplified dynamics. A more general safe set validation step can be implemented by looking up pre-calculated reachable sets obtained through Hamilton-Jacobi-Isaacs analysis.
3. The current A* search in space-time is slow, and does not provide real-time performance. This can be sped up by using alternative search algorithms.

We are currently working on an extension of this work. Stay tuned!

Citation

If you find our code useful in your own work, please cite our associated paper:

Z. Zhang and J. F. Fisac, “Safe Occlusion-Aware Autonomous Driving via Game-Theoretic Active Perception,” Robotics: Science and Systems (RSS), 2021.

@INPROCEEDINGS{Zhan-RSS-21, 
    AUTHOR    = {Zixu Zhang AND Jaime F Fisac}, 
    TITLE     = {{Safe Occlusion-Aware Autonomous Driving via Game-Theoretic Active Perception}}, 
    BOOKTITLE = {Proceedings of Robotics: Science and Systems}, 
    YEAR      = {2021}, 
    ADDRESS   = {Virtual}, 
    MONTH     = {July}, 
    DOI       = {10.15607/RSS.2021.XVII.066} 
}