MarineCore
Autonomy software for
the deep ocean.
MarineCore is the command & control platform for autonomous deep-sea vehicle — navigation, AI autonomy, sensor arrays, payload bays, and comms modeled as a running FastAPI service, not a static simulation.
Everything the console needs to run a mission.
Physics-grounded modeling and a live operator console — the same logic that would drive a real deployment, exercised continuously by an automated test suite rather than demoed once and left alone.
Physics-based propulsion & avoidance
Thruster and hydrodynamic modeling drive obstacle avoidance directly from real vehicle dynamics, not scripted paths.
Terrain-following depth control
Commanded depth is continuously clamped against live bathymetry, holding altitude above the seafloor as terrain rises and falls.
Live command console
A browser dashboard streams mission status, telemetry, and vehicle health in real time over WebSocket.
Seafloor mapping planning
Operators plan survey coverage with live trackline length, duration, and energy-budget estimates before committing.
Manual fly-by-wire
An operator can take direct heading/depth/speed control from autonomy at any moment, with an explicit handoff protocol.
Offline JavaScript fallback
If the WebSocket link drops, the console keeps running a full client-side fallback simulation rather than going dark.
Layered safeguards, not a single kill switch.
Every safety behavior below is enforced in code and covered by the automated test suite — including the honest boundary cases, not just the paths that make the demo look good.
Bathymetry-clamped depth
Commanded depth can never be issued below live seafloor readings — the safety margin holds regardless of what’s requested.
Forward-hemisphere avoidance
Obstacle detection with hysteresis reacts to what’s ahead, without flip-flopping avoidance on and off at the trigger boundary.
Layered power management
Autonomous responses stage at 30% battery (surface & recharge), 20% (return to base), and 8% (emergency ascent).
Autonomous decision logging
Every autonomy decision is logged with a timestamp and severity level — an auditable record, not a black box.
Manual/auto handoff protocol
Explicit, unambiguous control transfer between the pilot and the autopilot — no silent fighting over the stick.
Passive-sensing-only payloads
All four defense/ISR payload modules are restricted to passive sensing by design — enforced by a regression test, not just policy.
Every module below is exercised by a real pytest suite — run before every release, not written once and left to rot.
- Vehicle physics — speed/turn-rate/depth clamping, the real-unit battery & solar model
- Navigation — waypoint capture, obstacle avoidance triggering and hysteresis
- Mission planning — the phase state machine, mapping-area math, expanding-square SAR route
- Acoustics — exact TDOA position recovery, graceful degradation under timing noise
- Formation control & distributed consensus — bounded drift under sustained comms loss
See MarineCore run a mission live.
A 20-minute walkthrough of the command console, the autonomy stack, and how the test suite backs every claim on this page.