ScenicRules: An Autonomous Driving Benchmark with Multi-Objective Specifications and Abstract Scenarios

[ Paper | Documentation ]

ScenicRules is an autonomous driving benchmark designed to evaluate systems against complex, multi-objective specifications in diverse driving contexts. It integrates the following key features:

• Multi-Objective Specification: Supports the formalization of conflicting driving objectives and explicit priority relations using the Rulebook framework.
• Abstract Scenario Representation: Leverages the Scenic probabilistic programming language to model driving contexts in an expressive, compact, and interpretable manner.

Repository Status

We are actively building out the tutorials and documentation for the repository. Stay tuned!

Installation

We recommend installing ScenicRules within an isolated Python virtual environment to prevent dependency conflicts. You can create and activate a new virtual environment using venv as follows:

python -m venv venv_scenic_rules
source venv_scenic_rules/bin/activate

Once your virtual environment is active, clone the repository and install the package in editable mode:

git clone https://github.com/BerkeleyLearnVerify/ScenicRules.git
cd ScenicRules
python -m pip install -e .

Authors

ScenicRules was originally designed and implemented by Kevin Kai-Chun Chang, Ekin Beyazit, Alberto Sangiovanni-Vincentelli, Tichakorn Wongpiromsarn, and Sanjit A. Seshia.

Citing ScenicRules

If you use ScenicRules in your work, please cite the following paper:

@inproceedings{chang2026scenicrules,
  title={{ScenicRules}: An Autonomous Driving Benchmark with Multi-Objective Specifications and Abstract Scenarios},
  author={Chang, Kevin Kai{-}Chun and Beyazit, Ekin and Sangiovanni-Vincentelli, Alberto and Wongpiromsarn, Tichakorn and Seshia, Sanjit A},
  booktitle={IEEE Intelligent Vehicles Symposium (IV)},
  year={2026}
}