Real-Time Obstacle Avoidance for Posture Prediction

Collision avoidance in digital human modeling is critical for design and analysis, especially when there is interaction between the avatar and his/her environment. This paper describes a new algorithm for obstacle avoidance with optimization-based posture prediction. This new approach is motivated by a need for decreased computational time and increased fidelity for modeling and analysis of collision avoidance tasks. Posture prediction is run in an iterative loop while conducting collision detection to dynamically update collision avoidance constraints. It is shown that this approach is substantially faster than the basic method involving a fixed number of sphere-based avoidance constraints with a single optimization/posture-prediction run. The method is demonstrated using an upper-body virtual human model in a cab setting.