Anthropic's AI agent surpasses humans in controlling a robot dog: a breakthrough in the physical world
Anthropic's large-scale Project Fetch experiment has reached a new, radically higher level. While in August 2024 the neural network only assisted engineers, the Claude Opus 4.7 model has now fully taken control of a robot dog, completing setup and programming tasks 20 times faster than the best human teams.
During the second phase of testing, the AI worked almost autonomously, under minimal researcher supervision. The model independently connected to video sensors and LiDAR, wrote a manual control program, created a route monitoring system, and configured an object recognition algorithm. This was achieved without implementing specialized algorithms for hardware control — the progress turned out to be a side effect of general language model scaling.
The numbers speak for themselves
The results are impressive. Opus 4.7 proved to be 18 times faster than a team using older AI versions and 37 times faster than humans working without chatbot assistance. Moreover, the neural network wrote code that is 10 times more compact than human-written code — fewer lines, higher efficiency.
However, there were limitations. Despite successfully guiding the robot to its goal, the model failed to gently push a ball to a specific point. This requires complex real-time feedback, where humans still maintain an advantage.
A look into the future
Anthropic is confident that the industry is entering an era of "physical AI agents." In the near future, neural networks will be able to work with standard tools and equipment as effectively as they do today with software code. Notably, on June 13, the company suspended access to the Fable 5 and Mythos 5 models due to a U.S. government directive on export controls, highlighting the strategic importance of these developments.
Expert opinion: Anthropic's breakthrough is not just a demonstration of speed, but a signal that AI is beginning to master the physical world. However, difficulties with precise manipulations remind us that the path to fully functional robot assistants is still long. The key question is whether the industry can scale these successes without losing control over safety.