Character animation using reinforcement learning and imitation learning algorithms

Abstract

Real-time character animation for gaming and film industries is challenging and achieving production-ready quality is the hardest part. Managing time and resources also plays a vital role here. Animation through marker-based motion capture is quite a tiresome process that requires costly motion-capture suits, multiple cameras, and a large amount of storage space to store all the animation. In order to make advancements in the field of animation, AI can help us manage our time and resources as well as achieve high-quality animation. In this paper, we propose a model that aims to generate real-time character animation for biped locomotion in Unity ML agents using Reinforcement learning and Imitation learning algorithms. We first evaluate the training with solely the state-of-the-art RL algorithm, PPO. Then we analyze the combination of Imitation learning algorithms BC and GAIL in conjunction with PPO. We further discuss the comparison between the two training datasets and show that our model is able to generate animations in real-time avoiding all the tedious work and large databases. We demonstrate that this approach will result in a good amount of data compression making it effortless while maintaining the quality.