From policy to possibility

Real robot experiments.
One inference harness.

Deploy policies, explore supported inference strategies,
and turn real-world runs into reproducible experiments.

Explore ManiMux ↓

Product animation · Synthetic RoboGUI footage
Recorded UI preview · No hardware execution

01 / Built to connect

Policy × Runtime × Embodiment

Your stack.
Without the silos.

Connect supported embodiments, policies, and inference strategies in one experiment workflow.

01

Embodiment

The physical connection.

Dual-arm integration with deployment recipes and RoboGUI visualization.

02

Policy

The model behind the motion.

Pi05 deployment recipes, with model serving separate from robot execution.

03

Inference

The rhythm of execution.

Overlap inference and execution with asynchronous action chunking.

Explore each integration above. Availability depends on the model, adapter, and local station; selections do not certify a compatible combination. Experimental hardware awaits physical validation.

View support catalog ↗

02 / Close the loop

From intention to a saved experiment

A run is just
the beginning.

Bring deployment, robot execution, and human judgment into the same experiment workflow.

AGENT-ASSISTED DEPLOYMENT

03 / See it in context

Visible robots. Visible execution.

The interface
meets the world.

Physical camera views, robot visualization, and execution information, brought together in RoboGUI.

Different stations. Shared evidence.

We’re collecting continuous recordings with the physical robot, model identity, and live inference visible together.

Explore the project ↗

Open source. Open possibilities.

Let’s make
the next move.

ManiMux

An open-source harness for real-world
manipulation inference and experiment management

Paper link forthcoming

ManiMux product film