A Synthesis of Automated Planning and Model Predictive Control Techniques and its Use in Solving Urban Traffic Control Problem

Most desired applications for planning and scheduling typically have the characteristics of a continuous changing world. Unfortunately, traditional classical planning does not possess this characteristic. This drawback is because most real-world situations involve quantities and numeric values, which cannot be adequately represented in classical planning. Continuous
planning in domains that are represented with rich notations is still a great challenge for AI. For instance, changes occurring due to fuel consumption, continuous movement, or environmental conditions may not be adequately
modelled through instantaneous or even durative actions; rather these require modelling as continuously changing processes. The development of planning tools that can reason with domains involving continuous and complex numeric fluents would facilitate the integration of automated planning in the design and development of complex application models to solve real world problems.
Traditional urban traffic control (UTC) approaches are still not very efficient
during unforeseen situations such as road incidents when changes in traffic are requested in a short time interval. For such anomalies, we need systems that can plan and act effectively in order to restore an unexpected road traffic situation into a normal order. In the quest to improve reasoning with continuous process within the UTC domain, we investigate the role of
Model Predictive Control (MPC) approach to planning in the presence of mixed discrete and continuous state variables within a UTC problem. We explore this control approach and show how it can be embedded into existing, modern AI Planning technology. This approach preserves the many
advantages of the AI Planning approach, to do with domain independence through declarative modelling, and explicit reasoning while leveraging the capability of MPC to deal with continuous processes.
We evaluate the possibility of reasoning with the knowledge of UTC structures to optimise traffic flow in situations where a given road within a
network of roads becomes unavailable due to unexpected situations such as road accidents. We specify how to augment the standard AI planning
engine with the incorporation of MPC techniques into the central reasoning process of a continuous domain. This approach effectively utilises the strengths of search-based and model-simulation-based methods.
We create a representation that can be used to capture declaratively, the definitions of processes, actions, events, resources resumption and the structure of the environment in a UTC scenario. This representation is founded on world states modelled by mixed discrete and continuous state variables.
We create a planner with a hybrid algorithm, called UTCPLAN that combines both AI planning and MPC approach to reason with traffic network
and control traffic signal at junctions within the network. The experimental objective of minimising the number of vehicles in a queue is implemented to validate the applicability and effectiveness of the algorithm. We present
an experimental evaluation showing that our approach can provide UTC plans in a reasonable time. The result also shows that the UTCPLAN approach can perform well in dealing with heavy traffic congestion problems, which might result from heavy traffic flow during rush hours.