Marlin V3’s hull is a 9.5-inch diameter acrylic tube, 25.875 inches long, and sealed by two anodized 6061-T6 aluminum endcaps.

The front endcap is sealed to the end of the tube with 3M DP420 epoxy, and the front camera dome is sealed using a face o-ring. An aluminum collar was added to the back of the tube, so that the back endcap seals to aluminum instead of casted acrylic, ensuring seal integrity.

The back end cap uses o-rings 3% smaller than normal specification and was given a 10-degree chamfer to ease insertion and extraction of the electronics rack. Four spring-draw latches on the rear of the hull help keep the back endcap seal watertight by preventing the endcap from sliding out of the tube. Thirteen waterproof SubConn connectors interface between Marlin V3’s internal and external electronics, and can be quickly detached and reattached without disrupting the watertight seal. Two aluminum mounts attach the hull to the horizontal plane of the submarine.

Marlin V3 maneuvers using eight BlueRobotics T200 thrusters, chosen for their high thrust and efficiency. Four are arranged horizontally at 45-degree angles and control strafe, yaw, forward, and reverse movements, while the other four are oriented vertically and control depth, pitch, and roll.

The frame is built from 6061-T6 aluminum components for high strength and relatively low weight. Sheets and L-pieces are machined using CNC machines and standard industrial tooling machines and then anodized with a type II coating. Modular Panels were also integrated to easily adapt the frame to changes that need to be needed. These panels can easily be printed or machined to fit our new needs. The top plane has five while the front and back legs each have one giving us a total of seven spots to mount components. With the 3D-printed modular panels, all of our new parts like grabbers and torpedoes can easily be added in multiple orientations and locations on the submarine creating more possibilities.

Marlin V3 runs three main code-bases at the same time. Thalassic is the higher level program that the Jetson executes, which controls sub trajectories, handles vision detections, and directs the AUV through the course. Maritime is the lower level program that is run on our Raspberry Pi Pico microcontroller, which connects to our thrusters, pressure sensor, and kill switch. Pelagic is our secondary low level program that runs on our STM32 Black Pill, acting as a sensor stack and interfacing with auxiliary sensors useful for telemetry but not required for proper vehicle function. While Thalassic sends motor commands, maritime send the appropriate thrusts to each thruster over PWM.

To monitor the submarine, we employ a web graphical user interface (GUI), that displays camera image feeds, the sub’s position and orientation in the water, and a remote control interface used when collecting images.

To control Marlin V3 with increased precision, the high-level control program implements a cascade PID loop, with position and velocity controllers. The output of the position controller on six axes, one for each degree of freedom, serves as the velocity setpoint. The output of the velocity controller or the velocity error, then becomes the force and torque setpoints on each degree of freedom. This force vector for the overall AUV is mapped to thrusts for each individual thruster using a Quadratic-Programming Thrust Allocator, directing the submarine to correct its position and orientation error and converge on its waypoint.

To navigate through the course, Marlin V3 integrates DeepSeeColor, YOLOv11, and the Scale-Invariant Feature Transform (SIFT) to ensure robust target identification. As a preprocessing step, DeepSeeColor, a model-based underwater color correction algorithm, mitigates water-column attenuation and backscatter, significantly improving the reliability of downstream vision tasks. YOLOv11, trained on a combination of synthetic and real data, is then utilized to generate regions of interest (ROI) bounding boxes around mission elements. Finally, SIFT is applied within these bounding boxes to extract robust, scale- and rotation-invariant keypoints, enabling the precise spatial localization required for complex tasks such as the torpedo board.

The simulator remains an integral tool for the software team, as it allows for instant debugging and validation of new navigation and vision detection code. By utilizing the Stonefish simulation engine and designing the latest competition’s props according to their released specs, we can simulate competition settings to develop mission and motor control code to traverse the course. It serves as an invaluable tool, as it allows in-water pool testing sessions to be used primarily for validation of approaches that have already been tested in the simulator.

Mission generates a list of goals to complete throughout the competition run. Each goal contains a set of instructions, location, and point value. Based on the location of both the sub and the goal, we can compute an estimated time to completion for each goal, and visit the goals in an optimal fashion, maximizing our point value. Once we select a goal to complete, mission tells the corresponding vision function to start returning observations.