robots that learn at the edge of
land and water.
Brackish Labs builds physical AI for tidal flats, shorelines, and shallow water — the substrate most robotics still skips.
Most robots are built for one substrate. The interesting work is at the seam.
Quadrupeds are designed for dry ground. ROVs stay submerged. Almost nothing handles the meter of shoreline where footing turns to silt, where tide resets terrain every six hours, where a sensor has to survive mud, salt spray, and shallow water within the same hour.
The boundary is also where most coastal economic activity actually happens — aquaculture, port work, flood response, wetland survey — and almost none of it is automated.
Brackish Labs builds the perception, control, and hardware for machines that work there.
Four problems we think are worth solving in the next three years.
amphibious locomotion
Gait policies that transition between dry ground, soft mud, and shallow water without a mode switch. We treat the substrate as a continuous variable, not three discrete regimes.
perception under chop
Visual and inertial state estimation when the horizon line is wet, moving, and partially reflective. Standard SLAM assumptions break the moment a wave crosses the field of view.
salt-tolerant hardware
Long-duration actuators, sensors, and power systems that survive repeated saltwater submersion. The boring half of the problem and the half that decides whether anything ships.
field-conditioned learning
Closed-loop training on real coastal sites instead of simulators. We think the wet boundary is too non-stationary to sim-to-real cleanly, and we want to find out where that assumption breaks.
Hardware we are building or adapting.
Most of this is early. We will publish platform details as each one reaches field readiness.
A quadruped sized for intertidal work. Sealed joints, replaceable foot pads, ankle-deep wading.
A tethered surface-and-shallow-water platform for river mouth survey. Carries a small payload bay for third-party sensors.
Where we go to break things.
We prefer to publish site-level data and failure modes alongside the work, not just the highlight reel.
- intertidal mudflatsgait failure modes on soft, saturated substrate
- rocky shorelinesfoothold planning on wave-wet rock
- tidal marsheslocomotion through dense vegetation in standing water
- harbor edgesperception against reflective, moving water surfaces
We default to publishing.
Datasets, failure logs, and reference code will appear here as they are ready.
multimodal recordings from tidal field sites
what broke, where, why
how we evaluate field robots
land × water.
substrate is the work.
For research collaboration, field site access, or hardware questions — get in touch.