Automatable Evaluation Method Oriented toward Behaviour Believability for Video Games

摘要

Classic evaluation methods of believable agents are time-consuming becausethey involve many human to judge agents. They are well suited to validate workon new believable behaviours models. However, during the implementation,numerous experiments can help to improve agents' believability. We propose amethod which aim at assessing how much an agent's behaviour looks like humans'behaviours. By representing behaviours with vectors, we can store data computedfor humans and then evaluate as many agents as needed without further need ofhumans. We present a test experiment which shows that even a simple evaluationfollowing our method can reveal differences between quite believable agents andhumans. This method seems promising although, as shown in our experiment,results' analysis can be difficult.