Awkward movements of game characters, AI solvers?

As the game gives players the freedom to explore the complex digital world, the CG characters also move naturally and interacting with all the elements there is becoming increasingly difficult. The biggest problem is the unnaturalness between pre-programmed movements and movements.

To prevent this awkwardness, the research teams are trying to use AI and deep learning to move realistic game characters almost like real humans. In order to express motions such as walking, running, and jumping as realistically as possible, game developers capture real human movements and reflect them on digital characters. However, according to the research team, this is faster and looks better than manually animating the character, but it is impossible to shoot motion to fit everything how the character interacts with the digital world.

Game developers try to plan as many possibilities as possible. But in the end, you have no choice but to rely on software. The player’s gaming experience is compromised as the character’s continuous movements can feel exaggerated or unnatural.

The University of Edinburgh and Adobe Research researchers have found a new way to address this. It will be announced during the ACM SIGGRAPH conference held in Brisbane, Australia in December, and it will solve the awkwardness of animation that the game currently has by using neural network capabilities through deep learning.

This is like creating a deepfake video. When creating a deepfake image, the neural network first learns all the imaginable facial expressions using a database of tens of thousands of facial images photographed from various angles. This process takes time, but it allows you to automatically create realistic looking face replacements.

This study uses a similar approach, but instead of training a neural network from a facial database, it uses a series of digitized movements captured from actors on the sound stage for learning. For best results, you need an extensive database of actions an actor picks up, climbs somewhere, sits on a chair, etc. Once this is done, the neural network can adapt what it learns to fit almost any situation and environment while generating natural appearances and movements. Therefore, there is no awkwardness between the actions of the character walking towards the chair and slowly rotating the body or sitting. This is because it intelligently connects these movements and animations.

Complementing motion-captured with this technology can help reduce the file size and the amount of data that needs to be processed and shared. This technique can aid in complex interactions of game characters. Related information can be found here .