Evolving Behaviour Tree Structures Using Grammatical Evolution

Abstract

Behaviour Trees are control structures with many applications in computer science, including robotics, control systems, and computer games. They allow the specification of controllers from very broad behaviour definitions (close to the root of the tree) down to very specific technical implementations (near the leaves); this allows them to be understood and extended by both behaviour designers and technical programmers. This chapter describes the process of applying Grammatical Evolution (GE) to evolve Behaviour Trees for a real-time video-game: the Mario AI Benchmark. The results obtained show that these structures are quite amenable to artificial evolution using GE, and can provide a good balance between longterm (pathfinding) and short-term (reactiveness to hazards and power-ups) planning within the same structure. 1 Evolving Behaviour Trees for Game Playing Control The objective of this chapter is to demonstrate how Grammatical Evolution (GE) [1] can be used to evolve control structures for agents that interact in highly dynamic environments. An example of this application is the control of Non-Player Characters (NPCs) in computer games. These are highly dynamic environments, where NPCs must be able to react efficiently and effectively in previously unseen scenarios. In the concrete case of real-time games, these agents must be able to perform actions in a limited time budget, typically in just a few milliseconds. These actions must tackle both dynamic (enemies, power ups, moving structures) and static (level

Grammar-based cooperative learning for evolving collective behaviours in multi-agent systems

“…As such, GE can be identified as a more logical alternative that has the unique ability to utilise a CFG to control and restrict the solution structures to a desired syntax. This approach is increasingly being used in evolving behaviours of single agent systems [38,39,40] as well as homogeneous agent systems [8,11], due to its ability to generate syntactically correct solutions through the evolutionary process with reduced manual intervention. However, evolution of heterogeneous groups of agents has not been explored in a comprehensive manner with grammar-based approaches.…”

Exploiting abstractions for grammar‐based learning of complex multi‐agent behaviours

“…To address the challenges related to learning complex problems with direct approaches, abstractions are used as a way of decomposing the problem into simpler subproblems. In contrast to the traditional application of GE which has been tested in domains such as robotics and simulations 46,47 and gaming environments, 48,49 this paper investigates novel modifications to the GE model by exploiting modular attributes in the complex tasks to help distribute the learning load over time and/or multiple sublearners. The issues related to convergence at suboptimal solutions are expected to be overcome with the use of abstractions that decompose complex tasks into manageable subtasks, and encapsulating knowledge of learning these subtasks in separate sublearning modules.…”

Task Allocation in Multi-Agent Systems with Grammar-Based Evolution

“…GE has been studied with single-agent systems in the IVA domain such as for horse gait optimisation [17], and for controllers in gaming environments [20,21] and other multi-agent domains such as robotics and simulations [18,29]. Nevertheless, the existing GE models have not concentrated on distributed learning environments where IVA-based task allocation is concerned.…”