IDCA NewsAll IDCA News
10 Nov 2022
Code Should Be Generated by Robots, According to Google
Image credit: Google
As we enter the next decade of automation, countless big challenges still exist, and robot learning sits somewhere near the top. While humans have gotten pretty good at programming systems for specific tasks, one big question remains: What's Next?
New research demonstrated by a group of experts from Google's Artificial Intelligence event in New York City presented a result that allows robotic systems to write and reprogram codes.
The process is designed to reduce the hassle of reprogramming things as new information arises.
The company observed that existing research and trained models could prove helpful in implementing the concept. Furthermore, the entirety of that work can serve as foundational information when developing systems capable of creating code based on their encounters in an environment.
In a blog post, Google Research Intern Jacky Liang and Robotics Research Scientist Andy Zeng note:
"With CaP, we propose directly using language models to write robot code through few-shot prompting.
Our experiments demonstrate that outputting code improved generalization and task performance over directly learning robot tasks and outputting natural language actions.
CaP allows a single system to perform various complex and varied robotic tasks without task-specific training."
The system also relies on third-party libraries, APIs, and support for languages and emojis. The information accessible in those APIs is one of the current limitations.
The researchers observed, "These limitations point to avenues for future work, including extending visual language models to describe low-level robot behaviors (e.g., trajectories) or combining CaPs with exploration algorithms that can autonomously add to the control primitives."
Therefore, as part of this announcement, Google is making its source code available for reuse, releasing open-source versions through its GitHub site. Please remember, though, all of the caveats concerning early-stage research still apply.
Follow us on social media: