UST035

66 in the nose and another in the ‘tail’ to enable yawing and lateral thrust, plus four vertically oriented motors disposed in ‘wings’ at the corners of the AUV, which (depending on how they are combined) power pitch, roll and vertical thrust. The battery is a custom-built pack designed around maximising the ratio of kilowatts to litres, which is more important than kW/kg in the Sunfish’s case, as well as a 36 V power bus architecture and 18650 lithium cells (with about 3.6 Ah per cell). “Every year or so we revisit the battery design to try to upgrade it, using fresh research on the cell configurations, chemistries and other key battery technologies available,” Stone says. “But the constraint is that our packs all have to be flight-certified because we perform missions all over the world, so they need to go through judicious battery testing before we use them.” Software platform As is becoming standard on AUVs, the Robotic Operating System is used as the middleware for the Sunfish’s autonomy. Overlaid onto it are the various modules for high-level specific autonomous behaviours such as SLAM navigation and exploration, as well as some more classical state machines for performing comparably simple functions such as real-time sense & avoid. “Our autonomy is heuristically designed,” Stone says. “We’ve interviewed a lot of professional divers and explorers, and solving the problems with exploring caves, shipwrecks and similar complex enclosed spaces opens a lot of doors in terms of how algorithms can enable learning in unexplored environments.” The company’s algorithms are also modular, enabling prior information such as CAD models and maps on upcoming operational locations to be integrated into its mission planning software. Most of the code has been written in Python, C, or C++, with the first being chosen for ease of use when many Stone Aerospace programmers need to collaborate quickly on a new bespoke configuration of the AUV or another technology. SLAM navigation Stone Aerospace’s approach to underwater SLAM uses a combination of inertial measurements, multi-beam sonar references, acoustic Doppler velocity logging (DVL) and pressure-depth sensing. With these sensors installed, the Sunfish’s overall 3D SLAM process consists essentially of three broad stages or layers. The first is a dead-reckoning algorithm, for taking data inputs from the onboard IMU and outputting nominal estimates of the AUV’s position, orientation and velocity. “Although we can use MEMS IMUs, most of the time we go with a fibre- optic gyro [FOG] for accurate inertial sensing,” Stone says. “We can scale the inertial technology to meet a client’s safe accuracy requirement, but we don’t have a huge amount of compartment space for fitting an IMU. Litton – now Northrop Grumman LITEF – makes a FOG that’s very size-optimised, so we’ve used that for most missions. “At the most extreme, we’ve used a $500,000 ring-laser gyro from Honeywell to get really precise readings, for December/January 2021 | Unmanned Systems Technology Motors are incorporated around the hull for redundancy and complex manoeuvring through underground caves The company’s 3D SLAM navigation is optimised for exploring new environments upon launch, then autonomously deciding when and how to return for recovery