Safety evaluation of intelligent transportation systems using agent-based models

Recent intelligent transportation systems reduce traffic congestion and improve the overall efficiency of traffic networks. Such systems often involve complex algorithmic logic, large-scale traffic networks, and a high volume of data. For instance, a data-driven path recommendation engine [Gonzalez, 2007] for Illinois (USA) relays on a graph representation with 831,524 nodes and 1,048,080 edges. Their complexity is a challenge to ensure safety.

Agent-based modelling is a promising approach to build and analyse intelligent transportation systems because they can accurately capture intricate patterns in traffic flows. For example, the TransWorld [Wang, 2010] can simulate urban road networks to evaluate the effectiveness of smart traffic management systems using agents. However, there is little research applying agent-based models to assess the safety in intelligent transportation systems.

This project concerns agent-based modelling to evaluate the safety of intelligent transportation systems in a computationally efficient way. It includes four areas: modelling, safety evaluation, efficient computation, and case studies.

Modelling includes the design of an agent-based model for traffic networks. The model should capture assumptions and rules on the interactions between agents and the environment. Existing agent-based models usually focus on traffic efficiency. However, the model in this study will address safety-related issues, including weather, public lighting, and pedestrian jaywalking. Moreover, the model in this study should allow calibration algorithms to search and tune model parameters to fit existing data on traffic flows and accident records.

Safety evaluation includes simulation and model-checking. Simulation is the procedure that produces future scenarios based on an initial setting. Due to randomness in a traffic network, the simulation algorithm should generate different versions of future scenarios based on the same initial configuration. The distribution of future scenarios, especially those with traffic accidents, provides a measure of the safety of the intelligent system. Model-checking provides a way for hypothesis testing under uncertainty. In particular, as the traffic system is stochastic, Bayesian model-checking [Jha, 2009] will be adopted to check safety-related hypotheses to complement simulation.

Computational efficiency issues arise when the agent-based model is sophisticated. For instance, it is usually necessary to simulate the system for a considerable number of times to perform Bayesian model-checking and allow rare accidents to emerge. Concerning the computational efficiency issues, this study includes the design of accelerated software and hardware tools for calibration, simulation and model-checking on multi-cores, graphics processing units (GPUs) and field-programmable gate arrays (FPGAs).

Case studies include safety evaluation for typical intelligent transportation systems, including data-driven path recommendation [Gonzalez, 2007] and intelligent traffic light control [Wiering, 2004] using publicly available data from Transport for London (TfL). The agent-based models developed in this project should identify potential safety issues, especially traffic accidents caused by the deployment of these intelligent systems, with reasonable computational time and resources.

[Gonzalez, 2007] Gonzalez, H. et al. “Adaptive fastest path computation on a road network: a traffic mining approach.” VLDB 2007

[Wang, 2010] Wang, F.-Y. “Parallel control and management for intelligent transportation systems: Concepts, architectures, and applications.” IEEE Trans. Intelligent Transportation Systems 11.3 (2010): 630-638

[Jha, 2009] Jha, S.K. et al. “A bayesian approach to model checking biological systems.” Int. Conf. on Computational Methods in Sys. Bio., 2009

[Wiering, 2004] Wiering, M.A. et al. “Intelligent traffic light control.” Technical Report, UU-CS-2004-029, (2004).

Project ID




AI Planning, Verification