Anthropic's AI model Claude outperformed human engineers by 20 times in controlling a robot dog
A new phase of Anthropic's Project Fetch experiment has demonstrated an impressive breakthrough in artificial intelligence autonomy. The Claude Opus 4.7 model completed the full cycle of configuring and controlling a four-legged robot 20 times faster than the best teams of human engineers working with previous AI versions.
Evolution of Autonomy: From Assistant to Operator
In August 2024, Anthropic employees with no robotics experience attempted to program a robot dog with AI support. At that time, the neural network only acted as an advisor. In the new testing phase, Claude Opus 4.7 operated almost entirely independently, with minimal researcher supervision. The model independently performed the following key tasks:
- connected to video sensors and LiDAR;
- wrote a program for manual control;
- created a robot path monitoring system;
- configured an object recognition algorithm.
The key performance indicator is speed. Opus 4.7 turned out 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 was 10 times more compact than that of human teams, indicating a deeper understanding of the system architecture.
Side Effect of Scaling
The project authors note that this breakthrough was not the result of introducing specialized algorithms for controlling hardware. The progress in robotics turned out to be a side effect of the general scaling of language models. This confirms the hypothesis that universal AI systems can master physical tasks without additional training, simply by improving basic cognitive abilities.
Limitations: Physics Still Unyielding
Despite the impressive results, Claude still experiences serious difficulties with precise physical actions in real time. The model managed to guide the robot to its goal but failed at the task of gently pushing a ball to a specific point. Such operations require complex sensor and motor feedback, where humans still maintain superiority. As Anthropic aptly noted, "the robot dog never managed to fetch the beach ball."
Anthropic believes the industry is entering an era of "physical AI agents." In the near future, neural networks will be able to use standard tools and equipment as effectively as they currently work with software code. Recall that on June 13, the company was forced to halt access to the Fable 5 and Mythos 5 models due to a directive from the U.S. government under export controls, highlighting the growing geopolitical significance of these technologies.
Expert commentary from Cryptalist: This experiment is a vivid demonstration that we are moving toward a point where AI will not only analyze data but also independently control physical systems. However, the failure with the ball reminds us that from programming to real-world interaction, there is a vast distance. Investors should keep an eye on companies solving the "tactile feedback" problem — this will be the next major frontier.