About the project

The project addresses a class of problems concerning management of groups of entities (agents) featuring different levels of controllability, predictability and observability.

The essence of the problem is three-fold:

1. Incomplete knowledge regarding search domain, with detailed local data gathered through observation by plan executor.
2. Timely re-planning unfeasible due to fast real-time execution.
3. Executors disobedient or error-prone to certain extent.

Problems featuring the above characteristics cannot be fully addressed with classical planning methods. Therefore, many real-world situations remain unsolved or existing solutions are not satisfactory. Some examples of such problems include:

• coordination of mobile robots motion,
• micro-scale urban traffic management,
• vehicle route planning in urban conditions,
• network routing with strong QoS requirements.

The key results include development of decision support methods using robust multi-branch planning, generation of multi-branch plans in discrete and continuous search spaces using agent-based methods and exploitation of software agents for modelling and simulation of heterogenous entity groups.

The research fits well into pure research, i.e. research carried out to increase understanding of fundamental principles, as it assumes generalisation of the previously developed domain-specific methods towards domain-independent planning solutions. Special emphasis will be put on assuring interoperability of the devised multi-branch (robust) planning solutions for comparison with current state-of-the-art research being done on automated planning and scheduling, particularly heuristic planning, design of domain-independent heuristic functions, agent-based optimisation and robust execution.

The scientific results of the project will include:

• algorithms for generation and execution of multi-branch plans in discrete search spaces,
• agent-based methods for finding multi-branch plans in continuous search spaces
• multi-branch plan execution methods in environments containing agents featuring various levels of autonomy, controllability and observability,
• foundations for an advanced traffic management systems.

Even though the planned research is based on strong formal and scientific principles, it can be applied to solving numerous practical problems, including heterogeneous mobile robot group motion coordination, urban traffic management and advanced traffic simulation.

Some of these applications will be used for simulation-based and practical verification of the devised methods: modelling and management of urban traffic in micro-scale and mobile robot motion coordination. Application of software agents should allow for modelling and simulation of various vehicle control (driving) style. Such a model should also allow simulation of environments with co-existing vehicles featuring different levels of controllability and interaction of autonomous and human-driven vehicles. Performed simulations will include various levels of planner-driver communication richness and personalisation.

The developed algorithms will be verified experimentally using a distinctive research station created as part of the project. It will consist of two types of models of considered entities and their environment: mathematical and physical.

The mathematical model will be based on rigid body simulation methods and shall be implemented based on advanced robot simulation tools already available in the Institute. The tools supports modelling and accurate simulation of numerous groups of coexisting robots due to ability of parallel computation of rigid body dynamics.

The physical simulation model will consist of a group of vehicles, represented by mobile robots. Each robot will be equipped with local embedded computer, laser range-finder and cameras. The experiments will be performed in a dedicated laboratory. Initially in the Institute's existing mobile robot laboratory, and later in a dedicated laboratory organised for the purpose of the project.

Planned research includes design of a hardware vehicle, analysis of architectural issues and preparation of a vehicle prototype. These will be used for derivation of simulation model of a vehicle, capable of implementing the multi-branch entity group management and control methods developed in the earlier stages, based on the design and the prototype. Design, construction and assembly of approximately 20 vehicles is planned.

The meaning of the results for civilisation and the society include:

• Application of planning and search methods to decision-support rather than control systems. Classical planning assumes little autonomy or intelligence of the plan executor. In real-life problems, the planner often has vague knowledge about the state of the search domain. However, if we provided the executor with knowledge useful to make decisions on its observations, instead of providing it with a single plan, we could maintain its autonomy and make use of its intelligence. Examples of such problems include car navigation (with guidance of human drivers) and mobile robot control (with reasoning based on sensor data).
• Systematic verification of the influence of driving style on efficiency of urban traffic. Modelling of plan executor profiles using software agents and verification using logical and physical simulation models can provide useful results regarding the effects of driving style on the efficiency of individual travel and the performance of the entire transport system.
• Feasibility study regarding co-existence of entities featuring various levels of controllability, observability and predictability in a common environment. The simulation framework will cater for arbitrary characteristics of the entities regarding their controllability (including autonomy, susceptibility to errors and disobedience, etc.) and observability (explicit or implicit feedback from the executor to the planner). This relates directly to situations concerning human- and computer-controlled entities, be it interaction of human-driven and autonomous cars in a common road system, or cooperation between humans and robots in manufacturing or medical applications.
• A realistic framework for evaluation of road infrastructure solutions. Most classical approaches to road infrastructure planning rely on analysis of vehicle flow (e.g. using cellular automata or queue networks). The proposed approach can supplement these methods with more fine-grained analysis of various design subtleties and their effect on traffic efficiency.