Our Research

Machines that know who is speaking, where they are, and what matters — starting with hearing.

Flagship Research

Acoustic Awareness for Embodied Robots

Every affordable microphone array is planar and azimuth-only; everything three-dimensional is a $30,000 laboratory instrument. Robots therefore receive sound as a stream while receiving vision as a scene. Project Daredevil closes this gap at the sensor: geometry-agnostic 3D spatial decomposition over any microphone constellation, user-loadable ML models on a single edge DSP, and identity-conditioned attention routing that emits a live spatial awareness map — identity, class, azimuth, elevation, range, confidence, priority — instead of raw audio.

Provisional patent filed May 2026. Open-source reference design in development, in partnership with Ultimate Bots.

Morphological Acoustics

Passive printed structures — cones, chokes, pinnae — are computation executed in geometry at zero watts and zero latency. One quarter-wave design curve (f = c/4L) governs both vocal projection (2–4 kHz presence band) and elevation perception (6–16 kHz pinna cue band). We design and test 3D-printable acoustic organs that give ordinary microphone arrays elevation sensing for the price of a print.

Persistent Identity & Relational Memory

Awareness without memory resets every morning. We build the layer above the sensor: enrolled voiceprint identity, cross-session continuity, and attention that knows the difference between a family member, a guest, and a television. This is where Mira AI's long-term value lives — the modules stay open; the relationships are the product.

Legible Machines

Field-tested doctrine: a robot proves it can hear by moving, not by speaking. We study reflex acknowledgment (direction-of-arrival to gesture in under 300 ms, decoupled from cognition) and response-channel selection — gesture in a crowd, speech at close range — as the foundation of human trust in hearing machines.

Interested in our research?

Explore our latest publications, collaborate with our team, or join us in shaping the future of AI.

The Daredevil paper has open contributor slots — substantive contribution earns authorship.