A Dual-Arm Robot Clocks 1,000 Hours on a PLC Line
Lumos Robotics' MOS2 carries 50 kg across two arms in half a square metre of floor space, and its predecessor ran failure-free for more than a thousand hours inspecting PLCs at a Mitsubishi Electric plant.

Lumos Robotics launched MOS2 on 31 August, a heavy-duty mobile manipulator it describes as an AI worker rather than a demonstration platform — a distinction the accompanying numbers largely support.
The machine stands about 1.65 metres tall and occupies roughly half a square metre of floor space, sized to move down narrow factory aisles. Its two arms carry a combined payload of up to 50 kilograms, with rapid end-effector changes to switch tools between workflows. It combines visual and force sensing, which the company says reduces the pre-calibration normally required before deployment.
What separates this from the usual launch is where it is already working. MOS2 is deployed on Mitsubishi Electric's core PLC production lines, doing flexible quality inspection, box pushing, screwdriving and material handling. The first-generation MOS robots completed more than 1,000 hours of continuous failure-free operation at Mitsubishi Electric's Changshu factory. In the inspection task the robots autonomously identify products of different sizes and use coordinated dual-arm movement to perform six-sided, unobstructed surface-defect inspection — which is a neat use of two arms, since presenting every face of a part to a camera without a fixture is exactly the job a second hand exists for.
Mitsubishi Electric opened four scenarios to the company — quality inspection, assembly, sorting and material handling — and the two set up joint laboratories in Suzhou and Shanghai. In May 2026, Mitsubishi Electric Smart Manufacturing Technology (China) led Lumos' Series A1 and A2 rounds, taking cumulative funding above one billion yuan. A customer that becomes lead investor is a stronger reference than any case study.
Chief executive Yu Chao framed the metrics that matter to industrial buyers as task success rate, production cycle time, deployment time and return on investment — which is the right list, and notably does not include anything about general intelligence. The open question the release does not answer is what that 1,000-hour figure covers: continuous operation without a robot fault is not the same as operation without human intervention, and for anyone building a business case, the difference is the labour line.
Source: Lumos Robotics