SayCan
A robot task-planning research system published by Google in 2022, demonstrating that language models can decompose high-level semantic instructions into executable low-level action sequences. It is one of the representative works establishing LLMs as planners in robotic systems.
Timeline (last 90 days)
The idea of natural language as an agent interface is in the same lineage as works such as SayCan and Inner Monologue.
Representative works applying LLMs to robot task planning include SayCan (Google, 2022) and Code as Policies (2023).
Since 2023, researchers have begun using large language models such as GPT-4 and Claude as the decision-making layer for robots, with representative systems including SayCan, Inner Monologue, and RT-2.
Google's SayCan project, proposed in 2022, uses LLM knowledge as a value function to guide robot behavior, with action selection determined by multiplying LLM scores by the success probability of skill modules.
All Facts (4)
The idea of natural language as an agent interface is in the same lineage as works such as SayCan and Inner Monologue.
50%UnverifiedRepresentative works applying LLMs to robot task planning include SayCan (Google, 2022) and Code as Policies (2023).
50%UnverifiedSince 2023, researchers have begun using large language models such as GPT-4 and Claude as the decision-making layer for robots, with representative systems including SayCan, Inner Monologue, and RT-2.
50%UnverifiedGoogle's SayCan project, proposed in 2022, uses LLM knowledge as a value function to guide robot behavior, with action selection determined by multiplying LLM scores by the success probability of skill modules.
50%