Updating local feature models based on modifications to robot actions

摘要

PROBLEM TO BE SOLVED: To make an action feasible or to execute it accurately. A method, device, and computer-readable medium for determining and utilizing modifications to a robot action. Some implementations are intended to update the robot's local feature model in response to determining a human modification of the action performed by the robot. The local feature model is used to determine one or more features that most closely resemble the generated embedding, based on the embedding generated on the corresponding neural network model. Updating the local feature model in response to a human modification can include updating the feature embedding of the local feature model corresponding to the human modification. Adjustments to feature models (s) can quickly improve robot execution without the need for retraining of the corresponding neural network model. [Selection diagram] Fig. 1