Motion fields with deep reinforcement learning for real-time character animation

Character motion in games and animations often have high requirements of realism, aesthetics, and interactivity. For instance, in soccer simulation games, users control the players to move in different directions and perform actions such as passing and shooting. Modern data-driven approaches like motion fields provide convenient ways to synthesizing natural motions from a given database of motion capture data. In this work, we look to improve motion fields by leveraging deep reinforcement learning. The benefit to the partner organization is the development of a new technology that can potentially improve the quality of their entertainment products as well as gaining expertise in these upcoming technologies.

Faculty Supervisor:

Michiel van de Panne

Student:

Partner:

Electronic Arts Canada (Burnaby, BC)

Discipline:

Computer science

Sector:

Information and cultural industries

University:

The University of British Columbia

Program: