Two thrusters
12 V thrusters with bidirectional speed controllers, one per hull. Heading hold from an IMU, tested against a desk fan standing in for the current.
AR-Mini: the same rescue, 61 centimetres long
A functional demonstrator, not a load-bearing platform. It swims, finds a warm dummy, lifts him into a cradle and drives out of the water. Built at home in Sofia, September to December 2026.
Drag to turn. Point at a part for its name.
610 mm
hull length, printed in two 305 mm sections
4.7 kg
mass budget, 42 mm draft
2 + 2
thrusters and thermal cameras
20
missions planned for December
The four phases, small
Swim
Find by heat
Two thermal cameras
One looks forward from a mast 274 mm above the waterline. The other looks down at the cradle to confirm the dummy is in it.
Recover
Servo cradle
A U of 5 mm aluminium rod on a parallelogram arm, so the dummy stays horizontal from the water to the deck.
Shore
Tracked chassis
A tracked chassis hangs under the cross-beams and takes over when the hulls touch the bottom.
Hull
Two hulls, 610 × 100 × 120 mm, parabolic bow, printed in PETG with four perimeters. Each hull is two sections, each its own watertight compartment behind 3 mm bulkheads. No holes below the waterline. Twelve M12 nuts are the ballast.
Pesho
The test dummy is 19–20 cm long and weighs about 100–150 g. He floats head and shoulders out, like a person in a life jacket. He is heated to body temperature with hand warmers or a small bottle of warm water, checked with an infrared thermometer. His head has six vents because hand warmers suffocate in a sealed head.
Electronics
The v1 controller is an ESP32-S3 development board designed and made in Plovdiv. A PWM driver runs the servos, two H-bridges run the tracks, serial-bus servos move the cradle and pan the camera, and an IMU gives the heading. The two thermal cameras share an address, so they sit on two separate I²C buses.
Power and safety
Two 2,200 mAh 3S lithium packs, one per hull, in parallel. From the battery: a 30 A fuse, a 50 A kill switch, then a six-slot fuse block. A second kill line cuts every PWM output without going through the controller. Battery voltage is watched with a buzzer threshold, and a leak sensor shares the same input. The simulation said energy is what fails first; this is the answer.
Tests
The first run on open water was on 1 November, at Pancharevo, driven from the phone with one thruster and a rudder. In December, feature freeze, then twenty missions, each written down the same way as the simulation trials: result, time, and the reason if it failed.
How it is built
I use AI tools for design and code. Every physical step is mine: printing, assembly, soldering, measuring, the water tests, the photos. Failures are kept on purpose: bodge wires stay in the pictures.
CAD, PCB files, firmware and the bill of materials are in the build repository as they are made, with real dates.
Scale: 1:9 by the person it rescues (a 20 cm dummy for a 1.8 m adult). Measured by hull length against the 3.2 m vessel in the paper, it is nearer 1:5.
What broke so far
Kept, because this is how the robot got its shape.
- Seven thrusters that weren't thrusters: fixed-speed pumps, bare motors, surfboard kitsRule: the title must say thruster, a thrust figure and a bidirectional controller
- A mast that exists because of a swinging armThe arm's swept sector pushed every beam aft
- A kill switch hiding on an analogue pinIt shares a resistor ladder with the leak sensor
- Pesho's breathing holesHand warmers need air
- The chassis was upside down in our heads for a dayReading the manual fixed it
- SDA and SCL swapped in softwareTo avoid one crossing trace on the board
- A print file almost sent to the board factoryThe .scad is for the printer; the .kicad_pcb is for the fab
- 32 design-rule violations, 2 real errorsStarved thermal reliefs on two ground pads