Discovering Fatigued Movements for Virtual Character Animation

Virtual character animation and movement synthesis have advanced rapidly\nduring recent years, especially through a combination of extensive motion\ncapture datasets and machine learning. A remaining challenge is interactively\nsimulating characters that fatigue when performing extended motions, which is\nindispensable for the realism of generated animations. However, capturing such\nmovements is problematic, as performing movements like backflips with fatigued\nvariations up to exhaustion raises capture cost and risk of injury.\nSurprisingly, little research has been done on faithful fatigue modeling. To\naddress this, we propose a deep reinforcement learning-based approach, which --\nfor the first time in literature -- generates control policies for full-body\nphysically simulated agents aware of cumulative fatigue. For this, we first\nleverage Generative Adversarial Imitation Learning (GAIL) to learn an expert\npolicy for the skill; Second, we learn a fatigue policy by limiting the\ngenerated constant torque bounds based on endurance time to non-linear, state-\nand time-dependent limits in the joint-actuation space using a\nThree-Compartment Controller (3CC) model. Our results demonstrate that agents\ncan adapt to different fatigue and rest rates interactively, and discover\nrealistic recovery strategies without the need for any captured data of\nfatigued movement.\n