Acoustic levitation · phased ultrasound array

Control matter
with sound.

An acoustic levitation platform with an AI-based controller. It positions and shapes fields of particles, cells, and droplets using phased transducer arrays. Contactless. No moving parts.

MageHand is a general-purpose acoustic levitation platform: phased transducer arrays driven by a controller trained with reinforcement learning.

Acoustic manipulation has existed in research labs for decades. What has kept it from becoming a platform technology is control: every new task, material, or geometry requires long expert hand-tuning. There is no general-purpose controller.

Reinforcement learning changes that. A single system can adapt to new particle types, geometries, and tasks. It abstracts away the expert tuning, so lab workflows can be reconfigured overnight, not redesigned from scratch.

Array
phased ultrasonic transducers
Control
learned policy, field-level
Status
verified in simulation

Space & microgravity

In orbit, surface tension rules liquids and loose particles drift freely. Acoustic fields offer programmable control. No contact, no contamination.

Programmable chemical laboratory

Build small molecules in situ from basic components. Software-defined synthesis pipelines replace bespoke fluidic hardware per experiment.

Life sciences & lab automation

Label-free manipulation of living cells, microparticles, and liquid droplets for lab-on-chip, diagnostics, and drug delivery.

Acoustic manufacturing

Position dust and powder fields in free space. Additive manufacturing without nozzles, substrates, or binders.

Aerosol engineering

Concentrate dilute aerosol clouds into dense pockets. Increase detector signals to reveal otherwise unreadable measurements.

Filterless filtration

Acoustic fields concentrate and remove coarse dust and pollens from gas streams with no replaceable filter element.

Field-level control

We don't move one particle at a time. We shape the trajectory of everything in the field, simultaneously.

AI-native

AI can adapt to new materials and geometries. Every conventional approach needs trial and error re-engineering.

Simulation-first

Trained on GPU clusters across billions of physics timesteps. Proven before heavy investment in hardware.

The AI controller works in simulation. Hardware is next.

Interested in collaborating or investing? Reach out.