System(s) and method(s) of using imitation learning in training and refining robotic control policies

US12226920B2 · kind B2 · utility

Assignee

• Google LLC · US

Inventors

• Seyed Mohammad Khansari Zadeh · San Carlos, US
• Eric Jang · Cupertino, US
• Daniel Lam · San Francisco, US
• Daniel Kappler · San Francisco, US
• Matthew Bennice · San Jose, US
• Brent Austin · Munich, DE
• Yunfei Bai · Fremont, US
• Sergey Levine · Redmond, US
• Alexander Irpan · Palo Alto, US
• Nicolas Sievers · El Cerrito, US
• Chelsea Finn · Berkeley, US

Key dates

• Filing date: Aug 11, 2023
• Grant date: Feb 18, 2025
• Expiry date: Aug 11, 2043

Classification

• Technology area (CPC G): Physics
• CPC primary: G05B2219/40391
• WIPO field: Handling
• WIPO sector: Mechanical engineering

Abstract

Implementations described herein relate to training and refining robotic control policies using imitation learning techniques. A robotic control policy can be initially trained based on human demonstrations of various robotic tasks. Further, the robotic control policy can be refined based on human interventions while a robot is performing a robotic task. In some implementations, the robotic control policy may determine whether the robot will fail in performance of the robotic task, and prompt a human to intervene in performance of the robotic task. In additional or alternative implementations, a representation of the sequence of actions can be visually rendered for presentation to the human can proactively intervene in performance of the robotic task.