Google Research and Everyday Robots , Google’s research arm of Alphabet in the United States , announced joint research “PaLM-SayCan” to solve the ” Moravec Paradox ” using advanced language models and robot learning. did.
By adopting PaLM (Pathways Language Model) as a learning model executed by Everyday Robots ‘ helper robots running on Google’s campus, the robot’s performance is enhanced and it is possible to make it perform more complex and abstract tasks. Research.
Moravec’s paradox is that AI and robotics have a harder time learning motor skills and perceptions based on human instinct than reasoning based on higher intelligence.
Everyday Robots’ robots understand and execute simple commands like “Bring me a bottle of water”, but “I spilled my drink. Can you help me?” It is difficult for a robot to give an order such as “Can you bring me a snack that will make you vomit?”
For example, I spilled my drink . and apologize. In PaLM-SayCan, the robot interprets the user’s request to bring something to clean up a spilled drink and decides to take a sponge.
At a high level, PaLM proposes multiple approaches to the task based on language understanding, and robots also propose multiple approaches based on feasible skill sets. The integrated system cross-references the two proposals and identifies a more useful and achievable approach for robots.
According to Google, the PaLM-SayCan robot was able to plan the correct response 84% of the time and execute it 74% of the time for 101 commands.
”We are excited about this progress… Our experiments demonstrated the robot’s ability to complete abstract natural language instructions with a high success rate. The interpretability of PaLM-SayCan allows users We believe that it will enable safe interactions with robots and robots,”