Robotics / Physical AI

commissure supports the work of running robots built by other companies in real settings in Japan. We work on the problems that stand in the way: tactile data that vision alone cannot provide, and integration into existing operations.

Getting the most out of a robot by choosing the right tactile sensor

For: Companies and system integrators bringing tactile sensing to robots

The problem

Choosing a tactile sensor suited to contact-rich work, and wiring it into a robot control system, becomes a barrier to adoption — in expertise and in engineering hours alike.

Our approach

We take on the selection, drawing on research into human touch and on hands-on evaluation of many sensors, and make integration lighter with our middleware, HaptoConnect.

Products used

  • HaptoConnect

Projects

We develop tactile sensors ourselves, and from that position we have evaluated a wide range of tactile sensors side by side. Our selection work builds on that. （We publish Tactile Arena, a catalogue of tactile sensors for robots）

Contact

Teleoperation that conveys force, not just contact

For: Robotics companies and operators on site

The problem

In the last-mile work that is hard to automate, the precision of the operation and the operator’s cognitive load both become problems.

Our approach

We return touch in real time through shear stimulation of the skin. Camera-based hand tracking keeps working, while operability improves and cognitive load falls.

Products used

  • FeelFuse

Projects

We have demonstrated haptic feedback in teleoperation using a pottery robot. （Named parties pending confirmation）

Contact

Recording human tactile data as training data for robots

For: Teams building robot learning datasets

The problem

Vision-based capture tends to fail under occlusion during grasping. The quality of the demonstration feeds straight through into how well the robot performs.

Our approach

FMG on the forearm captures the intent of a movement from before the grasp, and combines with video into multi-modal data. HaptoAI and our SDK support the path from generating training data to implementation.

Products used

  • SenseFuse

Projects

A proof of concept is in progress. We draw on research using muscle activity as imitation learning data (sources to be cited separately). （Publication timing to be set after the proof of concept concludes）

Contact

Why commissure

  • Haptic display that keeps the hands free
  • FMG that captures the intent of a movement from before the grasp
  • Vendor-independent integration that connects equipment from any maker
← Back to Solutions Contact