Google wants to let robots write their own code
Google has released a open source which it says will allow robots to perform tasks by writing their own code in response to instructions written by humans.
The company has launched a website (opens in a new tab) to reveal “Code as Policies” (CAP), whereby prompts written in plain English can be interpreted into language model-generated programs (LMPs) written in Python code.
CAP is the successor of PaLM-SayCan (opens in a new tab), a similar project that allows a physical assistant robot to be controlled via commands in plain English. CAP promises to allow more complex tasks to be completed with greater precision, partly by allowing machines to write their own code.
Self-coding robot
In one blog post (opens in a new tab) Discussing the release of the CAP, Google Research Intern Jacky Liang and Research Scientist Andy Zeng describe the driving force for the technology and what it means for the future.
“What if, guided by humans, robots could automatically write their own code to interact with the world? […] With instructions in natural language, current language models are very adept at writing not only generic code but, as we have discovered, code that can control robot actions.”
But maybe not the time to throw it away laptop for programming Just yet. In the test, the Google researchers demonstrated simple commands with a similar structure. The experimental robot can “draw a 5cm hexagon around the center” [of a whiteboard]and “put the blocks in a horizontal line near the top” [of a square boundary].
In the attachment paper (opens in a new tab), titled “Code as Policies: Language Model Programs for Embodied Control”, the project team acknowledges that CAP is currently incapable of handling particularly abstract or complex commands or perceiving descriptions of orbitals. The team’s approach also doesn’t account for the impossible commands given through the CAP.
In theory, the open-source nature of Google’s implementation of “robot-centric” LMPs in Python could lead to much faster implementation of solutions to these problems. The CAP website also has release (opens in a new tab) via Github and an interactive test (opens in a new tab) via Google Colab to describe how the robot “writes” code in response to commands.