Paweł Gora

List of main publications and preprints

1. Explainability of surrogate models for traffic signal control

   Gora, Pawel, Bogucki, Dominik, and Bolum, M Latif

   2023

   Abs HTML

   Traffic signal control is one of the most important means of managing and optimizing road transport in cities. Developing a new generation of efficient traffic signal control algorithms is challenging and it is expected that methods based on AI may bring advantages. However, explainability and interpretability of these techniques are often limited and sometimes it might be difficult even for traffic engineers to understand the decisions taken by such systems. In this chapter we focus on explainability of one of the AI-based techniques used for traffic signal-control which employs evolutionary algorithms to find heuristically optimal signal settings for given traffic conditions, and surrogate models based on graph neural networks to evaluate the quality of signal settings much faster than by traffic simulations. The considered surrogate model is a graph neural network in which the topology of connections between neurons is built based on the topology of the road network graph (neurons correspond to intersections with traffic signals, or road segments between the intersections). Therefore, it can be suspected that analyzing behaviour of these neural networks may also bring better understanding of the urban road traffic and the spatio-temporal impact of traffic conditions on various intersections. In order to investigate the effects of input features on the output of the graph neural networks, we calculated the Shapley values and applied the Zorro method. Thanks to applying these techniques, it was possible to assess importance of intersections and their impact on the times of waiting on red signals in the considered areas on points from randomly generated datasets and on datasets with 500settings found using genetic algorithm (considered as close to local optima). It turned out that both methods produce quite consistent results. Thanks to them, it was possible to identify the most critical intersections in the road network topology which might be important from the traffic engineering perspective in tasks related to traffic signal control.

2. Procedia Comput. Sci

   Microscopic traffic simulation models for connected and automated vehicles (CAVs)–state-of-the-art

   Gora, Paweł, Katrakazas, Christos, Drabicki, Arkadiusz, Islam, Faqhrul, and Ostaszewski, Piotr

   Procedia Computer Science 2020

   Abs HTML

   Research on connected and automated vehicles (CAVs) has been gaining substantial momentum in recent years. However, the vast amount of literature sources results in a wide range of applied tools and datasets, assumed methodology to investigate the potential impacts of future CAVs traffic, and, consequently, differences in the obtained findings. This limits the scope of their comparability and applicability and calls for a proper standardization in this field of research. The objective of this paper is to contribute towards bridging this gap by providing a summary of the state-of-the-art literature review regarding microscopic simulation models for connected and automated vehicles.

3. Transp. Res. Proc.

   Designing urban areas using traffic simulations, artificial intelligence and acquiring feedback from stakeholders

   Gora, Paweł

   Transportation Research Procedia 2019

   HTML
4. Training neural networks to approximate traffic simulation outcomes

   Gora, Paweł, and Bardoński, Marek

   In 2017 5th IEEE International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS) 2017

   Abs HTML

   We present results of our research on training neural networks to approximate traffic simulation outcomes, such as total times of waiting on a red signal. We developed TensorTraffic software, based on a TensorFlow library, and trained neural networks on a dataset generated by simulating traffic on a realistic road network of Warsaw using Traffic Simulation Framework software. The goal of conducted experiments was to approximate the total times of waiting on a red signal on a region of Warsaw (Stara Ochota district), with the input to neural nets representing offsets of traffic signals on that region. In the presented research, we focused on investigating different neural network models and strategies of their training. We took into account different sizes of training sets, different numbers of neurons and layers, different parameters of dropout and learning rate, in order to reduce as much as possible time required to conduct experiments and apply this method in practice (e.g., time required to generate training and test sets for neural networks, time to train neural networks and time to make inferences on a new set), while preserving a sufficient accuracy of approximations, which may be especially important from a practical point of view Results show that it is possible to train neural networks able to approximate with a high accuracy (with an average error 1.18%) outcomes of traffic simulations. Moreover, TensorTraffic allows obtaining results of approximations a few orders of magnitude faster than by running simulations using microscopic traffic models and it is possible to achieve acceptable accuracy on a relatively small training set (consisting of 10240 elements). It means that the method can be potentially applied to different traffic analysis and transport planning tasks (e.g., to find suboptimal configurations of traffic signals) and time-consuming computer simulations applied nowadays for traffic analysis can be potentially replaced by neural nets supported by computations using graphical processing units (GPU). This innovation may significantly reduce time required to complete research and engineering tasks related to designing road infrastructure and analysing vehicular traffic, as well as enable developing better traffic management systems. As an example, we applied the method to the traffic signal setting problem and accelerated existing genetic algorithm, giving opportunity to evaluate much larger set of possible settings and find better traffic management strategies.

5. Transp. Res. Proc.

   Traffic models for self-driving connected cars

   Gora, Paweł, and Rüb, Inga

   Transportation Research Procedia 2016

   Abs HTML

   Self-driving and connected vehicles, communicating with one another (V2 V technology) and with the road infrastructure (V2I technology), are a subject of extensive research nowadays and are expected to revolutionize the automotive industry in the near future. The major goal of our work is to design a microscopic traffic simulation model for such vehicles, including a robust protocol for exchanging information. The question arises as to whether such communication system may efficiently improve travel quality while reducing the risk of collisions. For the purpose of our research we created and developed a simulation software. Our tool visualizes traffic flow for custom but simplified road maps. The transport infrastructure includes multiple junctions, optionally equipped with traffic lights, and roads with varying number of travel lanes. Each vehicle is assigned a fixed route leading to a randomly chosen destination point. Any decisions made by autonomous cars (regarding acceleration or turning maneuvers) are preceded by communication stages (retrieving necessary data, negotiations). In the paper we present fundamental concepts, assumptions and design of our model and simulation software, we also discuss potential issues relevant to our approach. As for the future work, we plan to implement our model in a large-scale agent-based traffic simulation software, Traffic Simulation Framework, so that further examination will be carried out for realistic road networks taken from the OpenStreetMap project. We also plan to apply machine learning techniques, so that self-driving vehicles, as well as traffic light controllers, will be able to learn how to develop the best strategy and by this way improve traffic safety and efficiency in atypical cases.