Issue 39 Unmanned Systems Technology August/September 2021 Maritime Robotics Mariner l Simulation tools focus l MRS MR-10 and MR-20 l UAVs insight l HFE International GenPod l Exotec Skypod l Autopilots focus l Aquaai Mazu

97 environments and with waterproofing are also high priorities,” he adds. “We can also use them to inspect the condition of boat hulls, to study coral reefs, and before long we’ll be using Intel RealSense cameras to dock our robot in the charging station we’re working on.” To survey and measure water conditions, the Mazu contains a sensor sonde around the lower-middle of its body that can integrate up to 12 plug-and-play environmental sensors (interfacing via I2C or UART) to track parameters including oxygen levels, pH, turbidity, chlorophyll levels, salinity, electrical conductivity and oxidation- reduction potential. Pieterkosky notes that five environmental sensors and two cameras are typically enough for usual operations in fish farms. “We generally have one forward-facing camera and one on an undercarriage-mounted gimbal to acquire clearer images of features such as coral beds or dam walls,” he adds. “Across our iterations we’ve used configurations with side fins, multiple tails and so on, but the key thing for data gathering is the kind of smooth downward descent that buoys can achieve, for different levels of readings. Combining hulls and sensor suites in different ways is important for experimenting how best to manage that in each environment.” For delivering live video data and environmental telemetry to the GCS while underwater, the Nammu robots deployed at Kvaroy were also equipped with a tether. The Mazu can also be configured this way, or with a 250 Gbyte storage drive, as well as a cellular data link for uploading its data upon surfacing. “Alternatively, if we’re working with a smaller fish farm, we’ll use a comparatively inexpensive localised wi-fi network – modularity is key for keeping costs low for customers,” Pinto adds. “Also, we can compress the video stream and use other techniques for data handling; it all depends on customer requirements.” Pieterkosky notes, “All these different sensors, cameras and data links are COTS products, not only to save costs but because we want to use systems that are readily available around the world. We look for systems that we or our customers can access and replace as easily as possible, to minimise disruptions to operations and avoid any retraining or reprogramming requirements. “Cerulean Sonar for example makes an integrated IMU-GPS-DVL that’s been critical to meeting our displacement target. We’ve seen no-one else offer such a tight integration of navigation sensors for UUVs, all via a single waterproofed cable and potted connector.” Autonomy and AI The main computer uses ROS as middleware to process sensor inputs and output control functions, including following pre-planned mission paths based on dead reckoning between GNSS updates and activating different sensors and movements at key survey points. “The AI hardware is an Nvidia Jetson Nano, a very cost-effective computer that we can run at different power levels to help maintain efficiency and endurance,” Pinto says. In addition to acquiring raw data, Aquaai is developing a range of software tools including machine learning to incorporate computer vision and predictive analytics into the actionable data they deliver. “Computer vision is great for things like spotting holes in fishing nets, which can allow fish to escape into the wild, or measuring coral growth and biodiversity,” Pinto says. “And predictive analytics can model and project how water quality across farms will change over time, depending on whether the problems our robots identify are addressed or not.” Pieterkosky comments, “In previous iterations of our robot we’ve also performed SLAM, using two cameras to gauge and measure velocity changes based on movements relative to underwater objects. “While there are plenty of past studies on all these AI technologies that we could learn from, 90% of it can’t be used in aquaculture surveys. We’ve therefore had to work from a blank sheet in many ways, including acquiring large data sets of images and videos with our UUVs so that we can train our software at different light levels and water qualities.” Future plans At the time of writing, aquaculture, scientific research and port operators remained the biggest markets for the Mazu. Digitisation is taking an increasingly strong hold over blue economy industries as they seek more affordable tools to achieve the data quality needed for smarter growth policies. This is spurring growing interest in Aquaai’s platform, and its inevitable future versions can be expected to draw further attention. Unmanned Systems Technology | August/September 2021 Mazu Battery-electric biomimetic UUV Length: 1.3 m Horizontal diameter: 0.3 m Maximum endurance: 8 hours Maximum speed: 5 knots Maximum payload capacity: 100 kg Maximum payload complement: 15 sensors (12 survey instruments, three cameras) Some key suppliers Acoustic sensors: Cerulean Sonar Additional sensors: Blue Robotics Pressure housings: Blue Robotics Thrusters: Blue Robotics Underwater cameras: Granfoo Subsea connectors: Granfoo Pressure domes: Doer Marine Specifications Aquaai Mazu | Digest