Robots Atlas>ROBOTS ATLAS
Robotics & Hardware

arm-pilot: a Quest 3 drives a MELFA arm and logs physical-AI data

Lady Robot26 October 2026 · 2 min read
arm-pilot: a Quest 3 drives a MELFA arm and logs physical-AI data

HoloLab joined Mitsubishi Electric's partner programme for MELFA robots and released arm-pilot. The software drives an industrial arm with hand movements tracked by a Meta Quest 3 headset. It targets two things: dangerous work and training-data collection.

Key takeaways

  • arm-pilot controls MELFA robots through hand tracking on a Meta Quest 3
  • HoloLab joined the MELFA Robot Partner programme as a software partner
  • The integration uses the real-time external control function built into MELFA controllers
  • Target scenarios: high temperature, hazardous substances, cleanrooms, high-mix low-volume work
  • HoloLab is recruiting manufacturing and logistics firms for pilot projects

A hand instead of a teach pendant

arm-pilot pulls the operator's hand position and orientation from the Meta Quest 3 headset. The arm follows the hand in real time. There is no trajectory programming and no teach pendant.

The vendor lists two features of arm-pilot as central

hand trackingHand position and orientation read from the XR headset. The arm follows the operator's hand in real time.
native MELFAControl through the real-time external control function built into MELFA controllers. A native integration, not a wrapper over the operator interface.

Teleoperation as a way to collect data

The business case is twofold. The first is remote work where a human should not be: high temperature, hazardous substances, clean rooms.

The second is more interesting. Every teleoperation session leaves a trace — a recording of a human performing the task. That recording is training material for Physical AI. Teleoperation stops being only a control method and becomes a production line for data.

Human
Operator in a Meta Quest 3
Software
arm-pilot
Robot
MELFA controller
Arm motionAllow
Data
Trajectory logged
Physical AI training set

That explains why non-standard, low-volume tasks are on the use-case list. Classic robot programming does not pay off there, and learning from demonstration data is worth the most.

Who is behind it

HoloLab is based in Tokyo's Shinagawa ward. Founded in January 2017, it has just under 94 million yen in capital and is led by Kaoru Nakamura. It specialises in XR, AI and 3D spatial data for industry, construction and infrastructure.

In December 2025 it launched Xynapse AI for Enterprise with Beyondge, combining smart glasses, robots and AI. MELFA Robot Partner is a Mitsubishi Electric programme covering deployment and maintenance. HoloLab registered in the simulators and analysis tools category.

Why it matters

That reframes the economics. Consumer headsets instead of dedicated hardware lower the entry barrier further, and a logged session has value even when the operator makes a mistake.

What's next

  • HoloLab is taking applications from manufacturing and logistics companies for proof-of-concept projects and arm-pilot demos
  • The company plans to roll out remote-control solutions together with Mitsubishi Electric and the programme's other partners

Sources

Share this article