{"id":794018,"date":"2021-11-16T08:00:33","date_gmt":"2021-11-16T16:00:33","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=794018"},"modified":"2021-11-09T11:52:30","modified_gmt":"2021-11-09T19:52:30","slug":"research-talk-evaluating-human-like-navigation-in-3d-video-games","status":"publish","type":"msr-video","link":"https:\/\/www.microsoft.com\/en-us\/research\/video\/research-talk-evaluating-human-like-navigation-in-3d-video-games\/","title":{"rendered":"Research talk: Evaluating human-like navigation in 3D video games"},"content":{"rendered":"
On the path to developing agents that learn complex human-like behavior, a key challenge is the need to quickly and accurately quantify human-likeness. While human assessments of such behavior can be highly accurate, speed and scalability are limited. The researchers address these limitations through a novel automated Navigation Turing Test (NTT) that learns to predict human judgments of human-likeness. They demonstrate the effectiveness of their automated NTT on a navigation task in a complex 3D environment. They investigated six classification models to shed light on the types of architectures best suited to this task, and they validated them against data collected through a human NTT. The best models achieve high accuracy when distinguishing true human and agent behavior. At the same time, the researchers show that predicting finer-grained human assessment of agents\u2019 progress towards human-like behavior remains unsolved. Their work takes an important step towards agents that more effectively learn complex human-like behavior.<\/p>\n