Work with omni/omni-live to extract animation, fine tune models to generate real time behavior and expression in response to stimulus and internal motivation of the character. Design, train and evaluate reinforcement learning and imitation learning algorithms for lifelike movement, gaze and posture, conditioned on the robot's affective and motivational state. Work with animators to convert a library of hand-authored reference motion into style-conditioned policies, using adversarial motion priors and related techniques, so that character can be authored rather than hand-tuned.
Create scenario suites, batch simulation, video capture, and automated critics that assess behavior reads correct to an observer. Run paired human comparisons with designers and human-factors partners, and report how well automated critics correlate with human judgment.
Job description provided by Google.
View the original posting
