Those who watched the Tokyo 2020 Olympics would have likely observed the nearly 2,000-strong drone swarm lightshow during the opening ceremony which produced some spectacular 3D images. Controlling such a large number of drones to be in specific places at specific times is more difficult and less reliable using traditional “central processing station” methods. Instead, ease of implementation and reliability of the swarm scales better when the drones can compute and communicate locally to converge to a desired global (swarm) behaviour. This method is known as distributed decision-making.
Traditional decision-making architectures for agent-based population are centralised, achieving complex tasks by piping information from all agents to a common processing centre from which actions are then delegated to each agent (Figure 1A). Whilst this architecture is relatively simple and allows for relatively fast decision-making, it is fragile with the entire population’s capability relying solely on the operation of the common processing centre. In the real-world, the operation of the common processing centre is not guaranteed and thus it is desirable to utilise a decision-making architecture that has greater redundancy. A distributed decision-making architecture allows a population (tens or hundreds) of agents to achieve global goals via local communication to their neighbours (Figure 1B). This removes the requirement for a common processing centre and instead relies on neighbouring nodes communicating locally to find consensus on a set of actions that will achieve a global goal. Therefore, the loss of any single agent in the distributed network has a negligible impact on the global decision-making capability of the population.
Figure 1: Centralised (A) and distributed (B) networks.
- Aims and Objectives
The overarching aims of this project are to learn about software engineering practices through the development of a software system that can test and compare different methods of distributed decision-making. Furthermore, there are three key objectives of this project:
- Create a software system for testing distributed decision-making methods.
- Experiment with different distributed decision-making methods.
- Compare the effectiveness of each method.
A hypothetical “Cowboys & Aliens” (Figure 2) battle scenario will be considered to evaluate the performance of various distributed decision-making algorithms. The scenario will progress within a virtual space with an army of friendly agents (cowboys) who are defending something from an army of unfriendly agents (aliens). The aliens know the location of the thing the cowboys are trying to defend and want to destroy it. Each cowboy has their pistol, and some may be on horseback, whilst the aliens have their lasers, and some may be in their spaceships.
Importantly, the cowboys can only communicate verbally to other nearby cowboys to coordinate neutralisation attempts on the attacking aliens (not all cowboys can communicate with all other cowboys). The effectiveness of the cowboy army will depend on how well they can coordinate themselves in a distributed manner.
Figure 2: A battle scene is shown in “Cowboys & Aliens” (Universal Studios, 2011).
Cowboys & Aliens. (2011). Directed by Jon Favreau. Universal Studios-Dreamworks II Distribution.
本网站支持 Alipay WeChatPay PayPal等支付方式
E-mail: email@example.com 微信号:vipnxx