Project Epsilon
Our fifth-generation autonomous underwater vehicle — purpose-built for RoboSub competition, engineered for precision at depth.
Built to dive deeper — and think faster.
Epsilon is Project Nebula's fifth and most capable underwater vehicle, designed from the ground up for autonomous deep-water operation. Every system — from power distribution to the vision pipeline — was engineered in-house by our eight-member team.
The vehicle's modular architecture allows rapid reconfiguration between competition tasks. A new aluminum pressure vessel, tight cable management, and a centralized compute stack make Epsilon the most reliable machine we've ever put in the water.
At the core of Epsilon's intelligence is a ROS2-based autonomy stack fused with a real-time computer vision system capable of detecting, classifying, and tracking mission objects at depth — all without a tether.
By the numbers.
Under the hull.
Every major system — designed, built, and integrated by our team.
The vehicle frame is assembled from schedule-40 PVC pipe with custom 3D-printed T-joint connectors, housing a central circular aluminum pressure vessel. A modular endcap system allows fast payload swaps. The manipulator arm provides active gripping for object-handling tasks.
The power system is built around a custom power distribution board managing ESC feeds, 5V/12V regulation for compute, and fused protection for all rail outputs. All cables route through penetrator glands with secondary redundancy sealing.
Epsilon's software core is a ROS2 Humble architecture on Ubuntu 22.04 running on an NVIDIA Jetson Orin NX. A hierarchical state machine drives mission execution while a hardware abstraction layer decouples all driver code from business logic.
The vision system uses an OAK-D stereo camera feeding a custom-trained YOLOv8 model for real-time underwater object detection. Stereo depth maps provide 3D localization of targets, feeding directly into the navigation controller for closed-loop visual servoing.
Cascaded PID loops handle depth, heading, pitch, and roll stabilization. An extended Kalman filter fuses IMU, DVL, and pressure data for robust state estimation. A thruster allocation matrix translates 6-DOF commands to individual motor outputs.
What Epsilon has to do underwater.
RoboSub 2025 mission tasks — solved autonomously, no pilot, no tether.
Epsilon's journey to the water.
Meet the team behind the machine.
Eight builders, one submarine. Learn about the people who designed, wired, coded, and dove Epsilon into existence.