Multiple Sync’d Self-Recording, Battery-Powered Cameras

A still or video camera provides an important context for a collected sample or recorded sensor data. But even with a wide-angle lens, nominally spec’d at 63°, only a small fraction of a 360° surrounding view can be imaged. Imagine, though, a panoramic view of the seafloor surrounding a lander with time synched images from multiple cameras.

Perhaps those cameras are arrayed in pairs, ocular distance apart, providing 3D views in multiple directions. With a stereo visual headset—also known as a stereoscopic virtual reality (VR) or mixed-reality (MR) headset—each camera’s image is displayed on dual independent displays, feeding a slightly different image to each eye. This creates realistic 3D depth perception and immersive spatial awareness.

Costs have likewise been a constraint. Many researchers struggle with small budgets and look for solutions using low-cost cameras such as GoPro, Raspberry Pi, or Arduino. (See Lander Lab #9, Marine Technology Reporter, Sept/Oct 2023.) Reduced size leads to reduced costs.

DeepWater Exploration (www.dwe.ai), a San Diego based company, has taken on these challenges and design aspirations.

Their application of artificial intelligence, edge processing, reduced power demands, and increased digital storage capacity has led to imaging systems able to deploy for entire seasons, making intelligent real-time decisions in situ, recording terabytes of usable data.

Computer vision, especially underwater, is only as good as the raw image feed. Distorted imagery, high latency (the time it takes to capture, compress, and write the file), and dark frames (light sensitivity) turn even the most sophisticated AI models into guessing machines.

For decades, the marine industry got by with “dry” cameras housed in waterproof enclosures or bulky industrial cameras placed behind thick dome ports which inherently warped images. That worked fine when a human pilot could compensate, but it’s a whole other obstacle to train a machine to understand distorted views.

When used by autonomous systems, those optical flaws compound fast. Distortion throws off visual odometry, and chromatic aberration hinders feature tracking and object detection. In a two-knot current, a rolling shutter camera sensor skews targets with the “jello effect.” Add 100-200ms of latency, and real-time navigation and obstacle avoidance becomes unreliable.

DeepWater Exploration (DWE) started because a handful of UCSD engineering students, all veterans of ROV and AUV competitions, hit a wall with off-the-shelf hardware. With industry IP cameras priced well beyond their team budget, they did what any self-respecting broke engineering student would do: build their own camera from the ground up.

In a 2-bedroom apartment in La Jolla, California, they assembled the first custom PCB, machined a housing shaped around it, and applied their now patented lens technology. This was the first working prototype of the exploreHD.

What started as a solution for their next competition immediately gained interest from commercial operators, research teams, and defense groups. Three years later, over 3,000 exploreHD cameras have now been shipped and deployed worldwide to 600+ organizations across 50 countries.

Today, DWE hardware ranges from 400-meter ratings down to 11,000-meter full-ocean depth. The stellarHD series came next for high-speed computer vision, followed this year by the explore3D stereo camera, pushing their capabilities into spatial perception beyond vision alone. Across every model, the core architecture remains the same as the original exploreHD.

Pain Points of Legacy Cameras

Autonomous subsea operations consistently hit five major optical pain points, each of which have been measured and repeated:

1. Geometric Distortion

Standard dome ports and poorly calibrated dry lenses routinely cause 8-18% trajectory error in underwater visual SLAM (Simultaneous Localization and Mapping), even after heavy software correction (ref: ISPRS Annals 2023).

In open water, a vehicle drifting off course is manageable. Under polar ice or in a deep trench, that compound drift can turn a multi-million-dollar asset into a permanent seabed feature.

DWE’s Aquagon wet hemispherical lens fixes this by removing the water-dome-air interface entirely.

2. Low-Light Performance

Below 600-800m, older STARVIS sensors, Sony's trademark for back-illuminated pixel technology used in CMOS image sensors, start to struggle. In low-light environments, successful feature matching drops to 20-25% or less (ref: Frontiers in Marine Science 2023).

Vehicles surfacing with poor imagery can mean lost science, or failed missions.

DWE utilizes back-illuminated Sony sensors paired with custom low-light pipeline tuning for clean color imagery.

3. Latency

Latency above 80ms drives collision rates above 60% during dynamic obstacle-avoidance. (ref: Science Robotics 2020 - the same latency vs reactive performance curve widely cited across IEEE and ICRA research).

Instead of pushing video through IP network stacks, DWE engineered their imaging pipeline around direct USB connectivity which keeps glass-to-glass latency under 65ms.

4. Power Draw

Legacy IP-based cameras carry extra circuit boards, full network stacks, and inefficient power circuitry that pull 8 to 15W continuous at full resolution.

DWE stripped out that extra hardware, even removing indicator LEDs on the internal PCB, to reduce active power draw down to 1W for exploreHD and 1.5W for stellarHD cameras. When idle, the cameras run off <0.001 W, allowing them to sleep for months until triggered to snap a frame.

Lower power from sensors means the difference between a 30-day lander deployment and a months-long run.

5. Rolling Shutter

Rolling shutter in realistic motion introduces 10-30% higher position error in visual odometry compared to global shutter (ref: Monocular Visual Odometry with a Rolling Shutter Camera, 2017).

In strong current or fast vehicle maneuvers, SLAM diverges and visual targets get smeared.

DWE designed the stellarHD series with global shutter at 60 FPS, paired with frame synchronization and strobe light triggers, to keep frames crisp during high-speed surveys.

Making Scalable Vision

Good optics are only part of the equation. Scaling them through a modular and open ecosystem is what unlocks new capabilities.

DWE cameras are tiny, about the size of a golf ball, meaning they can fit almost anywhere. You can clamp one to a manipulator arm for close-up views, attach them to a docking head for alignment, or line them along the perimeter of a payload skid to eliminate blind spots.

Software setup is simple. Similar to basic webcams, DWE cameras show up as standard UVC devices over USB 2.0. Plug them in and they work on Linux, Windows, or macOS without custom drivers or proprietary wrappers.

Furthermore, multi-camera setups do not multiply system complexity. You just add extra units as needed. For a self-recording deep-sea lander or long-endurance platform, a DWE multiplexer board lets you run up to seven cameras through a single Raspberry Pi, and logs all feeds to an SSD on battery power.

Upgrading a work-class ROV built with an IP-based architecture doesn’t mean a full overhaul. DWE bridged the USB-to-IP gap with their Subsea Vision Computer (SVC), which takes multiple USB camera feeds and outputs them across a single ethernet stream. This lets teams swap in low-power imaging without tearing out existing cable infrastructure.

DWE provides open-source software for camera configuration so engineers can route streams directly into ROS nodes or their existing software stacks.

Standardizing everything on modular USB allows for universal adoption, usable by student teams to commercial operators in the field.

Each Vertical Industry in Every Ocean

Science:

Deep-sea research requires sensor payloads that handle extreme depth without driving up project costs.

In a joint project between NTNU and Northeastern University (arXiv 2506.06476, June 2025), researchers mounted a three-camera stellarHD rig on the Minerva II ROV to generate real-time 3D SLAM maps around the Hercules shipwreck in the Trondheim Fjord.

At JAMSTEC (Japan Agency for Marine-Earth Science and Technology), researchers have adopted the titanium-housed exploreHD Challenger for their upcoming seafloor-docked vehicles.

Commercial:

Subsea asset inspection continues to shift away from heavy support vessels towards smaller, nimble field crews.

In 2025, Blue Robotics selected the exploreHD camera for “The Reef,” their curated marketplace of innovative products to enhance the BlueROV2 and BlueBoat.

Up in the North Sea, Tethys Robotics standardizes exploreHD cameras across its Mermaid ROV fleet for portable, autonomous wind-farm inspections.

During surface operations around the United Kingdom, Online Oceans utilizes swarms of the Scout USV for long-endurance ocean surveys, each equipped with an array of exploreHD cameras to monitor the horizon.

Defense:

Mass-produced autonomous surface and subsurface fleets require low-cost optical payloads designed for long deployments.

Seasats, based in San Diego, CA, uses multiple exploreHD cameras on their Lightfish and Quickfish USVs to give operators complete visual coverage above and below the waterline. Their missions range from five-month Pacific transits to long-endurance runs in arctic conditions.

HavocAI standardizes exploreHD cameras across its Rampage USV fleet, and utilizes the modular hardware to retrofit other existing hulls as they’ve added more autonomous vessels to their lineup since 2025.

Ocean Aero adopted DWE hardware on their fleet of Triton AUSVs, the uniquely hybrid platform capable of both long-term surface sailing and evasive submerging, which needed sensors that could survive repeated dives.

Future for DWE

As subsea vehicles become more affordable across the board, sensors remain one of the main cost bottlenecks. When sensor costs drop, entire platforms become cheaper, more accessible, and easier to deploy.

DWE’s current focus is expanding from single-camera video into stereo vision with the explore3D. Bringing real-time 3D perception down to standard USB payloads cuts the cost of spatial awareness, giving observation-class vehicles advanced survey capabilities traditionally reserved for high-end commercial systems.

Acknowledgements

Thanks to Jake Johnson, DeepWater Exploration, for his significant help in writing and editing this article. The story benefitted from his academic training and natural gift of composition. The authors express their gratitude to the entire DWE team, and especially the founders Jiajer Ho, Brandon Stevens, and Kunal Singla for paving the way of DWE’s journey. The DWE team thanks all friends and partners who’ve supported along the way, especially Kevin Hardy from Global Ocean Design who’s been there since the beginning.