This AI Research Presents FollowNet: A Comprehensive Reference Dataset for Modeling Car Following Behavior

Following another vehicle is the most common and basic driving activity. Following other cars safely reduces collisions and makes traffic flow more predictable. When drivers follow other vehicles on the road, the appropriate car following model represents this behavior mathematically or computationally.

The availability of real-world driving data and advances in machine learning have greatly contributed to the rise of data-driven car tracking models over the past decade. Models that rely on data to track a vehicle include neural networks, recurrent neural networks, and reinforcement learning. However, there are several limitations to the current body of research, as follows:

• For starters, car-tracking models are not yet well evaluated due to the lack of standard data formats. Despite the availability of public driving datasets such as NGSIM and HighD, it is difficult to compare the performance of the newly suggested models with existing ones due to the lack of standard data formats and evaluation criteria for models that follow the ones. automobiles.
• Second, the limited data sets in the current studies make it impossible to accurately represent the following behavior of cars in mixed traffic flows. Modeling a car’s following behavior with small data sets that do not consider autonomous vehicles has been the main emphasis of previous research, which comes at a time when both autonomous and human-driven vehicles share the road.

To solve these problems and create a standard data set, a new study by the Hong Kong University of Science and technology, Guangdong Provincial Key Laboratory of Integrated Communication, Tongji University and the University of Washington published a point of reference known as FollowNet. They used consistent criteria to extract car tracking events from five publicly available data sets to establish the benchmark. The researchers ran and evaluated five car-tracking reference models within the benchmark, spanning conventional and data-driven methodologies. They established the first standard for such behavior using uniform data formats to facilitate the creation of car tracking models. It can be difficult to handle various data structures and frameworks from different data sets, but their standardized car tracking benchmark takes this into account.

Two conventional models and three data-driven car tracking models (GHR, IDM, NN, LSTM and DDPG) are trained and evaluated using the benchmark. Five popular public driving datasets—HgihD53, Next Generation Simulation (NGSIM)54, Safety Pilot Model Deployment (SPMD)55, Waymo56, and Lyf57—compose car-tracking events that make up the proposed benchmark. The researchers analyze several datasets looking for patterns of car following behavior and basic statistical information. The results show the use of consistent metrics to evaluate the performance of the reference models. In particular, the Waymo and Lyf datasets show that the Car following cases occur in mixed traffic situations. The researchers did not include events with more than 90% static duration.

Collisions are still possible, even when data-driven models achieve lower spacing MSE than classical models. The development of models that follow cars with zero collision rates and fewer spacing errors is desirable. It would be beneficial to include collision avoidance capabilities to make data-driven models more practical and safer for real-world use. All cars are assumed to exhibit consistent and similar behavioral patterns at the proposed benchmark. Realistically, however, driving habits can differ significantly depending on the driver, vehicle, and traffic conditions. As a result, creating adaptive algorithms and representative data sets that cover a variety of driving styles, behaviors, and traffic situations is essential to include driving heterogeneity in car tracking models.

The researchers suggest that future data sets should incorporate additional features to further improve the performance and realism of car-tracking models. For example, a more complete picture of the road environment can be achieved by adding traffic signs and data on road conditions. Algorithms can also take into account complicated relationships and provide better predictions by integrating data about nearby vehicles and their activities. Future data sets will be able to better reflect real-world driving scenarios with the use of these additional data sources, enabling the creation of car tracking algorithms that are both robust and effective.

Review the Paper and ai-LAB/FollowNet” target=”_blank” rel=”noreferrer noopener”>GitHub. All credit for this research goes to the researchers of this project. Also, don’t forget to join. our 33k+ ML SubReddit, 41k+ Facebook community, Discord Channel, and Electronic newsletterwhere we share the latest news on ai research, interesting ai projects and more.

If you like our work, you’ll love our newsletter.

<!– ai CONTENT END 2 –>