Overview
We are very excited to provide the research and practitioners community with the following downloads for both data and code:
FLOW: a deep reinforcement learning framework for mixed autonomy traffic
FLOW is a traffic control benchmarking framework. It provides a suite of traffic control scenarios (benchmarks), tools for designing custom traffic scenarios, and integration with deep reinforcement learning and traffic microsimulation libraries. The corresponding download, available at the following URL, provides open-source source code for the community to use, a few starter scenarios, and benchmarks with a leaderboard system for researchers to benchmark their algorithms against others from the community. To run the framework, one needs to either download SUMO, an open-source traffic microsimulation software freely available, or have a license from AIMSUN, a commercially available microsimulation software. If you use the FLOW software, we kindly request that you refer to the following paper:
"Flow: Architecture and Benchmarking for Reinforcement Learning in Traffic Control." C. Wu, A. Kreidieh, K. Parvate, E. Vinitsky, A. Bayen. arXiv:1710.05465 [cs.AI].
Mobile Century Data
The Mobile Sensory Data was collected on February 8th, 2008, as part of a joint Berkeley-Nokia experiment funded by the California Department of Transportation. It contains half a day of GPS tracks sampled down to one GPS point every three seconds on the I-880 corridor in California. The data represents the movement of vehicles on the freeway and enables the reconstruction of traffic flow along the corridor. If you use this data set, we kindly request that you refer to the following article:
Evaluation of traffic data obtained via GPS-enabled mobile phones: the Mobile Century field experiment[ResearchGate]. J.C. Herrera, D. Work, X. Ban, R. Herring, Q. Jacobson and A. Bayen, Transportation Research - Part C, 18, pp. 568–583, 2010, doi: 10.1016/j.trc.2009.10.006
Implementation of an LWR Solver
The Lighthill-Whitham-Richards Partial Differential Equation (LWR PDE) is a seminal equation in traffic flow theory. It is widely used as a traffic model for freeways. The package proposes a sample implementation of a solver using the Lax-Hopf method. This is an alternate approach to the classically used Godunov scheme, which is common in numerical analysis. If you use the toolbox, we kindly ask that you refer to the following articles:
Lax-Hopf based incorporation of internal boundary conditions into Hamilton-Jacobi equation. Part I: Theory. C. Claudel and A. Bayen. IEEE Transactions on Automatic Control 55(5) pp. 1142-1157, May 2010, doi: 10.1109/TAC.2010.2041976
Java Toolbox for Data Assimilation with Scalar Conservation Laws
The corresponding toolbox includes the Java implementation and manual for a toolbox that can provide solutions to a Riemann across a function. The package proposes a Java implementation of several advanced filtering techniques for data assimilation with a macroscopic traffic flow model discretized using the Godunov scheme and initial conditions corresponding to Riemann problem. Graphical interface enables the compression of different algorithms, such as the ensemble Kalman filter (EnKF) and the extended Kalman filter (EKF). If you download this toolbox, please refer to the following article:
On sequential data assimilation for scalar macroscopic traffic flow models. S. Blandin, A. Couque, A. Bayen and D. Work, Physica D: Nonlinear Phenomena, 241(17), pp. 1421-1440, Sep. 2012, doi: 10.1016/j.physd.2012.05.005.