Abstract
We are interested in performing dynamic tasks with humanoid robots. Using a model-based controller to perform these tasks can expose modeling errors that are not apparent when performing slower or less difficult tasks. To successfully perform a dynamic task with the Sarcos Primus humanoid, we augmented a model-based policy-mixing controller with automated model adaptation techniques. We conducted experiments to show that this augmented controller can balance the Sarcos Primus humanoid on an unstable seesaw platform in the presence of unexpected disturbances and significant model errors.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have