Runtime Motion Adaptation for Precise Character Locomotion

Character animation is a critical component of games and interactive applications. Recent data-driven methods rely on motion capture to generate high-quality real-time locomotion models for movement such as walking or running. However, these methods do not easily accommodate more subtle movement such as side steps or turning, necessary for the character to reach a precise position and orientation. These stepping movements are often needed when the character must interact with the environment, such as positioning in front of a chair so a sit-down animation can be played appropriately. This work addresses the problem of generating such motions in the low-data regime. Our goal is to create an expressive locomotion system able to reach a precise goal in the vicinity of a character. Our method takes as input raw motion capture data, and automatically converts it to a sequence database. At runtime, our system selects the most appropriate sequence and adapts it to very precisely reach the target. Furthermore, our system has a negligible performance impact, which makes it suitable for use in AR applications and video games.