Universal Robots and Anthropic have tested the Model Hardware Standard (MHS), a research-preview specification for AI agents to safely operate physical equipment, on a live cell of Universal Robots cobots. Given only plain-language instructions, an agent running on Claude Opus 4.8 discovered four separate cobots, set them up, and coordinated them as a single cell, handing payloads between machines with no pre-built integration code. Universal Robots calls the result "the strongest evidence we have seen that an agent could eventually shoulder a real part of that work."
The test ran on PolyScope X, Universal Robots' newer, API-first control system, with the agent operating through Claude Code. MHS gave the agent a standardized way to find each cobot on the network, read what it could do, and issue commands, the same driver-based approach Anthropic has used to connect liquid handlers, plate readers and microscopes in lab settings. The agent explored the cell on its own, worked out how the four robots related to each other, and could be asked mid-run to optimize the handoff sequence. Nothing about the connection between agent and hardware was hardcoded in advance.
Safety stayed outside the agent's reach. Devices declare their bounds, interlocks and emergency stops directly in MHS, and the agent inherits and operates within those limits while Universal Robots' own safety architecture keeps final authority underneath. Engineers who have spent weeks commissioning multi-robot cells will care about the claim that follows: Universal Robots says integration, not the robot itself, has been the real bottleneck to automation at scale, and this test targets that cost directly.
MHS remains a limited, application-only research preview built on a provisional specification, not a generally available product. Universal Robots says the same approach should extend to autonomous mobile robots, though it has not tested that case yet. Elsewhere in the preview, Doosan Robotics is testing MHS on its own robotic arms specifically for automated quality assurance and multi-robot coordination, one of several manufacturing and lab partners, including Tecan, QIAGEN and Danaher, working with the standard before Anthropic open-sources it.
