Amazon Web Services, with Strands Robots.
Hugging Face, with LeRobot.
Raspberry Pi.
Automata.
Universal Robots.
These partners will collaborate on safety assessments and the development of best practices for AI systems operating physical equipment.
The new standard from Anthropic aims to simplify the connection between AI agents, robots, and equipment used in experiments
Amazon Web Services, with Strands Robots.
Hugging Face, with LeRobot.
Raspberry Pi.
Automata.
Universal Robots.
These partners will collaborate on safety assessments and the development of best practices for AI systems operating physical equipment.
The Model Hardware Standard incorporates a labeling system to describe the physical limitations of each piece of equipment. The information can include characteristics such as weight and reach of a robotic arm.
The labels can also specify adjustable parameters, measurement options, and safety limits. This data can be integrated into a reference file so that a model can quickly understand the characteristics of a device it was not trained on.
Anthropic is currently working with an initial group of scientific laboratories and advanced manufacturers during this stage. Among the participants mentioned by the company are:
Anthropic presented the Model Hardware Standard, a system that aims to connect AI agents with physical devices.
The proposal initially targets scientific laboratories and could reduce integration time between equipment from weeks or months to hours or minutes.
The Model Hardware Standard (MHS) is a set of standardized drivers created by Anthropic. Its goal is to enable AI agents to interact with different devices without developing a specific integration for each piece of equipment.
The system offers a common interface and a shared format for exchanging data between devices connected to a network. According to Anthropic, this eliminates the need to create a specific "translator" program between each component.
For now, the MHS is in a "research preview" stage and is primarily aimed at scientific laboratories.
The MHS can operate directly through commands and API code files, without the need to use AI models. However, its integration with the Model Context Protocol allows for controlling devices using natural language.
According to Anthropic, a model could reason through each stage of an experiment, modify parameters in real-time, and in some cases, recover from hardware errors without human intervention.
The company demonstrated an example of Claude adjusting a laser, reviewing the results through a camera, and repeating the procedure to automatically calibrate the system.
It also proposed a scenario in which AI controls a microscope, analyzes what it observes, and decides which area needs further observation.