Yaskawa puts Google DeepMind's Gemini Robotics ER 1.6 inside MOTOMAN NEXT to build an agentic industrial robot

Yaskawa Electric Corp. said on July 15, 2026 it had developed an agentic robot system with Google DeepMind, pairing its MOTOMAN NEXT AI robot with Gemini Robotics ER 1.6, Google DeepMind's generative AI model for robot reasoning. In the system the Google model acts as the robot's decision-making intelligence, assessing the situation and devising the sequence of steps, while MOTOMAN NEXT executes the work, according to Yaskawa's release.

Yaskawa did not disclose financial terms, launch customers or prices. It said specific applications and delivery timing would be announced once preparations are complete. The release was issued two days after Yaskawa reported a separate physical AI demonstration with SoftBank, and one day before it joined Fujitsu's physical AI initiative with FANUC and Kawasaki Heavy Industries.

What the system does

According to Yaskawa, the system lets a robot take an instruction such as sorting a set of components, work out the order of tasks itself and carry them out without detailed programming or teaching. If an object is dropped during transport, the robot can recover automatically. The system can also be linked to production management systems so that tasks flow from factory software rather than from a technician at a teach pendant.

The execution side relies on three capabilities that Yaskawa says come standard on MOTOMAN NEXT: machine vision that recognises the environment and object positions with high precision, path planning that calculates safe routes around obstacles, and force sensing that judges grip status and contact. Gemini Robotics ER 1.6 supplies the higher-level reasoning that decides what to do with those capabilities.

The SoftBank track

On July 13, Yaskawa and SoftBank said they had demonstrated a physical AI system that handles wire harnesses, flexible parts whose shape and position are hard for conventional robots to predict. The demonstration used a vision-language-action model, which generates motion from camera images and task instructions, trained with SoftBank's AI data centre GPU cloud. Yaskawa said the physical AI runs as a module inside its existing robot control system, and that the workflow automates data collection, model training, simulation evaluation and deployment on real robots. The work builds on an earlier partnership agreement between the two companies.

Yaskawa's numbers

The AI push comes as Yaskawa's robot business is under pressure. In the quarter from March to May 2026, consolidated revenue rose 10.6% to ¥138,982 million but operating profit fell 19.2% to ¥8,486 million, according to its July 10 filing. Robotics segment revenue grew 2.0% to ¥56,728 million while segment operating profit dropped 82.3% to ¥888 million, which the company attributed to production effects from a core system migration and restructuring costs in Europe. Yaskawa kept its full-year forecast of ¥580,000 million in revenue, up 7.0%, and ¥60,000 million in operating profit, up 26.8%. On Aug. 31 it launched the MOTOMAN-HC12 collaborative robot with a 12 kg payload, 1,410 mm reach and a newly developed torque sensor.

Rivals and Google's two partners

Google now works with both of Japan's largest robot makers through different units. FANUC announced in May a collaboration covering Google Cloud's Gemini Enterprise and the Intrinsic platform, said it participates in Google DeepMind's Gemini Robotics Trusted Tester Program, and on Sept. 11 unveiled a Gemini-based AI Welding Agent. Yaskawa's deal goes directly to a Gemini Robotics model for task reasoning. Kawasaki Heavy Industries, whose precision machinery and robot segment grew fastest of the three, has focused on its own physical AI centre in San Jose and on healthcare with Fujitsu.

The architecture divides labour in a way that is becoming common in robotics. A large multimodal model handles what is often called embodied reasoning, interpreting the scene and planning steps, while the robot's own controller handles the precise, safety-critical motion. That split lets Yaskawa keep its strengths in motion control and safety while borrowing reasoning it would struggle to build, and it means the same robot can be upgraded as Google releases newer models.

For global buyers, the significance is that high-mix tasks such as kitting, sorting and line-side handling, which have required extensive teaching, could be set up by instruction. For Google, a second Japanese robot partner broadens the installed base on which Gemini Robotics models run in factories.

The risks are concrete. Yaskawa has not published success rates, cycle times or a release date. Dependence on a Google model raises questions about latency, offline operation and data handling that factory customers will ask. Yaskawa is also running several AI tracks at once, with Google DeepMind, SoftBank and Fujitsu, while its robotics segment profit has nearly vanished, which could stretch engineering resources.

The next markers are Yaskawa's results for the June to August quarter, any announcement of applications and launch timing for the Google DeepMind system, and further VLA work with SoftBank.