BARK is a semantic simulation framework for autonomous driving. Its behavior model-centric design allows for the rapid development, training, and benchmarking of various decision-making algorithms. It is especially suited for computationally expensive tasks, such as reinforcement learning. A a good starting point, have a look at the content of our BARK-Tutorial on IROS 2020.
For whom it is: Python evangelists implementing python behavior models or ML scientists using BARK-ML for learning behaviors.
Bark is available as PIP-Package for Ubuntu and MacOS for Python>=3.7. You can install the latest version with
pip install bark-simulator. The Pip package supports full benchmarking functionality of existing behavior models and development of your models within python.
After installing the package, you can have a look at the examples to check how to use BARK.
| Highway Example | Merging Example | Intersection Example |
|---|---|---|
 |
 |
 |
For whom it is: C++ developers creating C++ behavior models, researchers performing benchmarks, or contributors to BARK.
Use git clone https://github.com/bark-simulator/bark.git or download the repository from this page.
Then follow the instructions at How to Install BARK.
To get step-by-step instructions on how to use BARK, you can run our IPython Notebook tutorials using bazel run //docs/tutorials:run.
For a more detailed understanding of how BARK works, its concept and use cases have a look at our documentation.
Example Benchmark is a running example of how to use BARK for benchmarking for scientific purposes.
- BARK: Open Behavior Benchmarking in Multi-Agent Environments (IROS 2020)
- Graph Neural Networks and Reinforcement Learning for Behavior Generation in Semantic Environments (IV 2020)
- Counterfactual Policy Evaluation for Decision-Making in Autonomous Driving (IROS 2020, PLC Workshop)
- Modeling and Testing Multi-Agent Traffic Rules within Interactive Behavior Planning (IROS 2020, PLC Workshop)
- Formalizing Traffic Rules for Machine Interpretability (CAVS 2020)
- Robust Stochastic Bayesian Games for Behavior Space Coverage (RSS 2020, Workshop on Interaction and Decision-Making in Autonomous-Driving)
- Risk-Constrained Interactive Safety under Behavior Uncertainty for Autonomous Driving (IV 2021)
- Risk-Based Safety Envelopes for Autonomous Vehicles Under Perception Uncertainty (Arxiv)
The BARK ecosystem is composed of multiple components that all share the common goal to develop and benchmark behavior models:
- BARK-ML: Machine learning library for decision-making in autonomous driving.
- BARK-MCTS: Integrates a template-based C++ Monte Carlo Tree Search Library into BARK to support development of both single- and multi-agent search methods.
- BARK-Rules-MCTS: Integrates traffic rules within Monte Carlo Tree Search with lexicographic ordering.
- BARK-MIQP: MINIVAN Planner based on MIQP for single- and multi-agent planning. Check out the build instructions.
- BARK-DB: Provides a framework to integrate multiple BARK scenario sets into a database. The database module supports binary serialization of randomly generated scenarios to ensure exact reproducibility of behavior benchmarks across systems.
- BARK-Rule-Monitoring: Provides runtime verification of Rules in Linear Temporal Logic (LTL) on simulated BARK traces.
- CARLA-Interface: A two-way interface between CARLA and BARK. BARK behavior models can control CARLA vehicles. CARLA controlled vehicles are mirrored to BARK.
If you use BARK, please cite us using the following paper:
@inproceedings{Bernhard2020,
title = {BARK: Open Behavior Benchmarking in Multi-Agent Environments},
author = {Bernhard, Julian and Esterle, Klemens and Hart, Patrick and Kessler, Tobias},
booktitle = {2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
url = {https://arxiv.org/pdf/2003.02604.pdf},
year = {2020}
}
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.
BARK specific code is distributed under MIT License.