UST035

67 Stone Aerospace Sunfish AUV | Digest example in a National Geographic exploratory dive we did.” A Kalman filter fuses these with added inputs from the DVL and pressure- depth sensors to update and correct the nominal estimates. This step also calculates the internal biases of the IMU’s accelerometers and gyroscopes – velocity information coming from the DVL is especially important to this aspect of the Sunfish’s SLAM. Stone adds, “We’ve worked extensively with Teledyne RDI on our DVL, as well as Imagenex for our multi-beam sonar. Most of that work has been pushing them to miniaturise their sensors to fit into the AUV; they’ve gradually done that and enabled us to incorporate their systems more easily.” The RDI Explorer DVL installed offers 80 m of range, while Imagenex’s DeltaT sonar has a range of 120 m with a 120- beam swathe. The range limits of DVL technology, as well as the concentration of multi-path reflections in underwater tunnels and near metallic structures, can reduce the accuracy of the velocity information it produces in the kinds of environments in which the Sunfish operates. To counteract these limitations, the dead-reckoning algorithm is designed to use individual DVL beam velocities, and is also structured to identify outlier beam readings by testing for the smoothness and plane-fit of consecutive measurements (rejecting those readings that fail such tests). If DVL readings are entirely unavailable for periods of time, an embedded model of the Sunfish’s dynamics is used to estimate drift and compensate for it in the algorithm’s outputs. The second SLAM layer is the perception sensing stage, referred to as the SLAM front end. The Sunfish begins gathering data on its surroundings using the DeltaT sonar, which gathers acoustic return amplitudes over time. That raw data is then processed to measure the locations of beam hits relative to the echo sounder, before collecting the locations over time and matching them to estimate how the vehicle has moved relative to the scanned environment. This level of processing is key, as the AUV will rotate at a fixed position while firing beam pulses in order to generate a broad point cloud of sonar measurements with the same origin. Additional algorithmic steps here include a function to compensate for vehicle dynamics by adjusting for time- of-flight delays, and a Kalman smoother for backwards-filtering of attitude estimates produced during the scan. Most forms of SLAM have used vision or laser sensors for this type of front-end sensing. However, the attenuation of light through water and the scattering from particles in it make cameras and Lidar unsuitable for subsea SLAM. The third and final SLAM layer is the back end, which stitches together the front end’s individual estimates of the AUV’s movement to generate a pose graph. This is a chart that estimates the vehicle’s trajectory within the real-time generated map, with estimates and constraints for its movement. On top of this localisation capability are additional layers of programming to enable autonomous exploration, to ensure that the Sunfish actively maps out the edges of the caves and passages in which it is deployed without needing a direct command link or preprogrammed set of waypoints. With the sonar scans detecting obstacles, the AUV will steer around obstacles and through open corridors to fill in previously unseen edges of its maps. If it reaches a dead end, it will calculate optimal routes back through spaces already explored, to continue seeking the unseen areas or return to its original launch point. “Our code itself is platform- independent: we’re not limited to using this system on underwater vehicles,” Stone says. “Our SLAM and exploration algorithms could be transferred to a UGV, a quadcopter, a spacecraft – all autonomous vehicles use reasonably similar technologies to figure out where they are and how they can manoeuvre. “With sufficient inputs, any of them can function using our behaviour-based software modules. It’s easy for us to use an environment-specific sensor like a camera or Lidar in place of the sonar, for example, so a UAV could explore a cave or a disaster area like a collapsed building, and map it out before anyone goes in, potentially saving them from serious harm.” Looking ahead For the immediate future, Stone Aerospace is continuing to optimise both the standard and customised variants of the Sunfish, as well as providing r&d and exploration services for a range of end- users and different environments. With a number of missions already completed in various locations here on Earth, the Sunfish also looks a good fit for exploring the alien environments on the moons of Jupiter and Saturn. Unmanned Systems Technology | December/January 2021 Sunfish AUV Dimensions: 161 x 47 x 20 cm Weight: 50 kg Depth rating: 200 m (upgradeable to 1000 m) Endurance: 10 hours Range: 4 km Payload capacity: 500 g Payload bay volume: 1.7 litres Some key suppliers Multi-beam sonar: Imagenex Doppler velocity log: Teledyne RDI Conductivity/temperature sensor: Neil Brown Ocean Sensors Ring-laser gyros: Honeywell Fibre-optic gyros: Litton (Northrop Grumman LITEF) MEMS IMUs: off-the-shelf Thrusters: in-house Specifications