Key facts
- Anthropic has launched Model Hardware Standard (MHS), a framework for AI agents to control physical devices.
- MHS enables AI agents to operate laboratory and manufacturing instruments, facilitating complex tasks like drug discovery experiments and laser calibration.
- The framework allows devices and AI agents to communicate across networks, aiming for autonomous, round-the-clock workflows.
- MHS is model-agnostic and works with any large language model (LLM), not just Anthropic's Claude.
- Anthropic is collaborating with device manufacturers to integrate MHS into new and existing products.
Anthropic has introduced the Model Hardware Standard (MHS), a new framework designed to allow artificial intelligence agents to operate physical devices. This initiative marks Anthropic's entry into the realm of "physical AI," aiming to bridge the gap between advanced large language models (LLMs) and hardware used in scientific research and advanced manufacturing.
The MHS framework is intended to enable AI agents to control laboratory and manufacturing equipment, such as microscopes and robotic arms, facilitating complex tasks and autonomous workflows. Anthropic stated that the MHS can help researchers streamline the often arduous process of creating custom software integrations needed for experiments, reducing setup time from weeks or months to hours or minutes.
According to Anthropic, MHS provides a common interface and data format for devices, allowing them to communicate across a network without requiring bespoke "translator" programs. While MHS devices can be controlled directly via command-line prompts and API code, integrating them with an AI model allows for natural language interaction, real-time parameter adjustments, and error recovery without human intervention.
Anthropic gave examples of an AI model using MHS to automatically calibrate a system by adjusting a laser and checking results via camera, or focusing a microscope and analyzing data to continue an experiment. The company also demonstrated Claude reasoning how to operate a robotic arm to pick up an aluminum can, even without specific training for that task.
The MHS framework includes a standardized tagging system to describe hardware's real-world constraints, such as physical characteristics, adjustable parameters, and safety limits. This information is integrated into a reference file that provides AI models with crucial details about devices they may not have prior training on.
Anthropic is currently working with a select group of scientific research labs and advanced manufacturers, including Amazon Web Services (Strands Robots), Hugging Face (LeRobot), Raspberry Pi, Automata, and Universal Robots, during an MHS preview period. These partners will assist in building safety evaluations and developing best practices for AI systems operating physical equipment. The company plans for MHS to eventually become an open-source and "agent agnostic" standard.