Case studies
Early-stage, and honest about it.
We'd rather show you one real, verifiable reference deployment than a wall of logos we can't back up. Here's what's been validated so far, more will be added as pilot deployments complete.
Simulated fleet, live flight environment
A multi-drone reference deployment running the full pipeline against a live PX4/Gazebo flight simulation, with a real semantic-segmentation perception stack under audit and real flight telemetry feeding the sensor-health signals. A confirmed fault produces a real veto decision and a real alert, not a log line.
Stated plainly: by default, the veto's physical PX4 command is withheld, not the veto itself, which is still fully detected, logged, and excluded from federated aggregation either way. The reason is a documented upstream constraint in MAVSDK itself, not specific to this codebase: its single-client connection carries both telemetry and action commands, so sending a hold can disrupt the telemetry stream it shares the line with. An explicit flag restores the physical command for a setup where that trade-off doesn't apply, and a dedicated, acknowledged, retried failsafe veto path already exists in the codebase for exactly this problem, it just hasn't been exercised against PX4/SITL yet, so we're not claiming delivery statistics for it that we haven't measured.
Gazebo PX4 SITL
TayuTech, a FinEst Centre for Smart Cities (TalTech) pilot
TayuTech is a publicly funded EU pilot programme (2025-2028) run by the FinEst Centre for Smart Cities under Tallinn University of Technology, equipping public service vehicles with modular sensor units for real-time urban monitoring, across pilot cities in Estonia, Ireland, and Croatia. Our SDK is being integrated into the road-infrastructure perception component, known as CitySense, the project's original name, running real-time road-surface-defect and manhole detection on camera footage from a moving vehicle, with the trust layer watching the perception model's outputs on real, unedited detections rather than a curated demo reel.
This is honestly an active technical integration, not a finished success story: the current milestone is correct, verified operation against real detection output, not yet a published detection-rate or fault-recovery result specific to this deployment. We'll update this page with real numbers once there are real numbers to share, not before.
TayuTech
Want to be the next one on this page?
We're looking for early pilot partners with a real perception stack in production or close to it.