AI agents enter the physical world: Claude Opus 4.7 controls a robot dog 20 times faster than humans
Anthropic's Project Fetch experiment has reached a fundamentally new level. While in August 2024, teams of human engineers with no robotics experience were only learning to program four-legged robots with the help of AI, now the Claude Opus 4.7 model performs this work almost autonomously — and with staggering efficiency.
The latest testing phase showed: Claude Opus 4.7 handles the full cycle of setting up and controlling a robot dog 20 times faster than the best human team that used the previous version of the model (Opus 4.1) a year ago. At the same time, the neural network operated with minimal researcher oversight — almost like an independent engineer.
What did the AI do?
The model took on the entire range of tasks that typically require the involvement of several specialists. It independently:
- connected to the robot's video sensors and lidar;
- wrote a program for manual control of the device;
- created a trajectory monitoring system;
- configured an algorithm for object recognition and classification.
The results are impressive: Opus 4.7 proved to be 18 times faster than teams using older versions of AI, and 37 times faster than humans working without chatbot assistance. Moreover, the generated code was not only faster but also more compact — its volume turned out to be 10 times smaller than human-written equivalents. This indicates a higher efficiency in the architecture of solutions found by the neural network.
Side effect of scaling
An important nuance: Anthropic did not introduce specialized algorithms for controlling the hardware. According to the developers, the progress in robotics became a side effect of the general scaling of language models. In other words, the smarter AI becomes at working with text and code, the better it adapts to controlling physical objects.
Limitations: physics remains a challenge
Despite the triumph in programming, Claude still faces difficulties at the stage of precise physical interactions. The model successfully guided the robot to the target but could not neatly push a ball to the desired point. This task requires complex real-time feedback — an area where humans still maintain superiority.
At Anthropic, they are confident: the industry is on the verge of an era of "physical AI agents." In the near future, neural networks will use standard tools and equipment as naturally as they work with software code today.
Cryptalist's comment: The Anthropic experiment clearly demonstrates that AI is ceasing to be just a "digital assistant" and is beginning to master the physical world. However, the gap between virtual programming and real-world motor skills remains significant. It is this "bottleneck" — the ability for fine tactile interactions — that will become the next frontier for the entire robotics industry.