Multi-agent algorithms with assignment strategy pursuing multiple moving targets in dynamic environments

Tools

Afzalov, A. ORCID: 0000-0002-1456-542X, 2022. Multi-agent algorithms with assignment strategy pursuing multiple moving targets in dynamic environments. PhD, Nottingham Trent University.

Preview

Text
Aziz_Afzalov_Final_Thesis_to IRep.pdf - Published version
Download (2MB) | Preview

Abstract

Devising intelligent agents to successfully plan a path to a target is a common problem in artificial intelligence and in recent years, attention has increased to multi-agent pathfinding problems, especially due to the expansion in computer video games and robotics. Pathfinding for agents in real-world applications is a defined problem of multi-agent systems, where pursuing agents collaborate among themselves and autonomously plan their path to the targets.

There are multi-agent algorithms that provide solutions with the shortest path without considering other pursuers and several of those use coordination. However, less attention has been paid to computing an assignment strategy for the pursuers and finding paths that collectively surround the targets. Comparatively fewer studies have been on target algorithms either. Besides, the multi-agent pathfinding problem becomes even more challenging if the goal destinations change over time. Existing solutions consider either a single target with moving capability or multiple targets that are stationary. The work presented in this thesis considers multiple moving targets in multi-agent systems. Therefore, the path planning problem for multiple pursuing agents requires more efficient pathfinding algorithms. In addition, when the target algorithms are improved for advanced behaviour with moving capabilities that smartly evade the pursuers makes the problem even harder.

The research reported in this thesis aims to investigate multi-agent search algorithms to address the challenge associated with pursuing agents towards moving targets within a dynamically changing environment. In multi-agent scenarios, agents compute a path towards the target, while these target destinations in some cases are predefined in advance. Thus, this research proposes to investigate a solution to the path planning problem by utilising heuristic algorithms as well as assignment strategies for multiple pursuing agents. Furthermore, a state-of-the-art moving target algorithm, TrailMax, has been enhanced and implemented for multiple agent pathfinding problems, which aims to maximise the capture time if possible until timeout.

The focus of this thesis is the investigation of the assignment strategy algorithms to coordinate multiple pursuing agents and explore pathfinding search algorithms to find a route towards moving targets. This will be achieved by dividing it into two stages. The first one is the coupled approach where the assignment strategy with a given criterion finds the optimal combination based on the current position of players. The second stage is the decoupled approach, where each agent independently finds its path towards the moving target. On the other hand, targets flee from pursuing agents using the specified escaping strategy.

The novel contributions of the research presented in this thesis are summarised as follows:

- A new algorithm is developed that uses existing assignment strategies, sum-of-costs and makespan, to assign targets, and then runs repetitive A* search until reaches the target.
- An enhancement is provided for a state-of-the-art target algorithm that takes smart moves by avoiding capture from all pursuers.
- To improve efficiency, six new approaches are investigated to find an optimal agent-to-target combination for target assignment.
- A novel multi-agent algorithm is developed which uses cover heuristics to maximise its coverage to outmanoeuvre, trap and catch moving targets.

The proposed pathfinding solutions and the results presented in this thesis demonstrate a significant contribution towards search algorithms in multi-agent systems.