Build a submarine that survives the basin

Why this stage exists

Nothing else in the pipeline exists without a camera that can sit two or three metres under the surface, hold depth, and still be reachable over Wi-Fi. Divers are slow and cold, and a boat-mounted camera cannot look at the basin floor.

How it was done

1. 1

   Start from an open-source design instead of a blank page

   The platform is based on ESP-DIVE, an open-source ESP32 RC submarine with a 3D-printed hull and a magnetic drive coupling, so there is no rotating shaft seal to leak. Building on a published design meant the effort went into waterproofing, camera integration, and field reliability rather than re-deriving a hull.

2. 2

   Put the computer and the camera in the same module

   A XIAO ESP32-S3 Sense acts as both the onboard controller and the camera. It can capture up to QXGA (2048 × 1536) through a clear dome port; for live first-person-view operation we stream SVGA (800 × 600) at about 5 frames per second, which is what the Wi-Fi link and the operator actually need.

3. 3

   Control depth with ballast, not with thrust

   Two piston ballast tanks driven by geared motors let the submarine dive, surface, and hold neutral buoyancy at a target depth. Steering is a single rear thruster with a rudder. Holding depth passively matters for image quality: a hovering camera produces frames the detector and the reconstruction can both use.

4. 4

   Keep the antenna out of the water

   Radio does not travel through water, so the hull is tethered by a 5 m coaxial cable to a floating antenna buoy. The buoy stays on the surface and carries the Wi-Fi link while the submarine works below it.

5. 5

   Decide what the submarine does not do

   Real-time YOLO inference is beyond the practical capability of the ESP32-S3, so the submarine stays a sensing and streaming platform and the detector runs on a host laptop. This kept power draw, heat, and firmware complexity inside the hull low.

Specifications

Onboard computer / camera

Seeed XIAO ESP32-S3 Sense

Capture resolution

up to 2048 × 1536 (QXGA)

FPV stream

800 × 600 at ~5 fps

Buoyancy

dual piston ballast tanks, geared motors

Propulsion

single rear thruster with rudder

Link

Wi-Fi via antenna buoy on a 5 m coax tether

Maximum depth

5 m

Control range

~50 m from the buoy

Endurance

~30 min per charge

Decisions and trade-offs

Where the code lives

What still bites

• 5 m depth and ~30 min of battery bound how much of a water body one sortie can cover.
• The tether that saves the radio link also limits how far the hull can wander from the buoy.