机器人CoRL 20222022
Do As I Can, Not As I Say: Grounding Language in Robotic Affordances
Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Daniel Ho, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Montserrat Gonzalez Angulo, Roelof Janssens, Jacky Kasberg, Johannes Keerthi, Sergey Levine, Chelsea Finn, et al.Google Research, Everyday Robots
摘要
We propose SayCan, a framework that combines the high-level reasoning capabilities of large language models with the low-level robotic affordances provided by pretrained skills. SayCan grounds language instructions in what the robot can actually do, using value functions to evaluate the feasibility of each proposed action. This approach enables robots to follow complex natural language commands while remaining grounded in their physical capabilities.
language groundingaffordanceslarge language modelrobotic skillsvalue functiontask planningSayCan