Virtual crowds typically rely on a limited set of motion-capture clips, which can lead to visible repetition and reduce the perceived diversity and realism of the crowd. This project will investigate whether Epic Games’ new AnimGen generative animation framework can be used to create more varied and individualised motion for virtual crowds.
The student will develop a crowd simulation in Unreal Engine in which the crowd system controls the high-level movement and interactions between agents, generating trajectories that account for navigation, spacing and collision avoidance. These trajectories will then be used to drive AnimGen characters, allowing the overall crowd behaviour to remain coherent while generating varied full-body motion for individual characters.
The project will compare traditional clip-based crowd animation with generatively animated crowds, potentially varying performer or motion style while keeping the underlying crowd trajectories constant. A perceptual experiment will investigate whether generative motion increases the perceived diversity, individuality and naturalness of the crowd and reduces visible repetition.
The project combines generative AI, character animation, crowd simulation and human perception, and would suit a student taking Real-Time Animation and a Machine Learning module, with strong Unreal Engine and programming skills.