Evolving controllable emergent crowd behaviours with Neuro-Evolution
Crowd simulations have become increasingly popular in films over the past decade, appearing in large crowd shots of many big name block-buster films. An important requirement for crowd simulations in films is that they should be directable both at a high and low level, and be believable. As agent-based techniques allow for low-level directability and more believable crowds, they are typically used in this field. However, due to the bottom-up nature of these techniques, achieving high level direct ability requires the modification of agent-level parameters until the desired crowd behaviour emerges. As manually adjusting parameters is a time consuming and tedious process, this thesis investigates a method for automating this, using Neuro-Evolution (NE). This is achieved by using Artificial Neural Networks as the agent controllers within an animated scene, and evolving these with an Evolutionary Algorithm so that the agents behave as desired. To this end, this thesis proposes, implements, and evaluates a system that allows for the low-level control of crowds using NE. Overall, this approach shows very promising results, with the time taken to achieve the desired crowd behaviours being either on par or faster than previous methods.