UST 031

78 In operation | Roboat Requesting a pick-up For a waste collection operation to begin, one or several boats could be requested via a smartphone app or routine collections at key locations could be established and automated. This will probably involve the Roboats navigating to the docking points where they are needed (using a combination of RTK-GNSS and object recognition) so that the end-user can place their refuse on the boat platforms, then confirming that they have done so. The researchers also envision the Roboats collecting refuse from waste containers integrated into quay walls or from floating bins. A sensor in each container would detect when its capacity has been reached, and alert the nearest Roboat ‘marina’ (likely to be set up and overseen by port or municipality authorities) to send one of the USVs to receive the waste and deliver it to the city’s central waste collection system. Navigating waterways The arrangement of the Roboat’s thrusters allows it to manoeuvre in any direction without needing a rudder. To steer to the left, the right throttle is increased, and vice versa to steer to the right. Also, to ‘park’ at a waterfront or give way to a water taxi, it can move laterally. “This lateral movement is enabled by the sideways-facing bow and stern thrusters,” says Deinema. “We control the bow and stern motor speeds in opposite directions to keep the boat stationary in the y axis but moving gradually in the x axis.” Autonomous navigation uses not just RTK-GNSS positioning but an array of sensors for obstacle detection and avoidance. The most important is currently a 16-beam 3D Lidar scanner for the smaller prototypes. Deinema adds that for the full-scale prototype they might upgrade to using 64- or even 128-beam Lidars, with one installed in the front of the vessel and another in the back to ensure 360 º of perception at all times. The Lidars will generate a continuously updating point cloud of the surrounding environment, to perform object recognition through the use of clustering and neural network-based classifiers to ‘remember’ the shapes of specific types of vehicles and objects, as well as recall their behaviours and dynamics on the water. Rather than gathering all this data before testing, AMS Institute and MIT are experimenting with Lidar-based simultaneous location and mapping (SLAM) to accumulate an offline map, which the Roboats are using as points of reference when returning through known areas. This is gradually establishing patterns of congestion, as well as increasing the AI’s ability to navigate through GNSS- denied areas such as tunnels, arches or the underside of bridges. In addition to navigating individually in this way to find waste collection points, the Roboats can move in a coordinated swarm-like network in order to make the best use of the limited space in the canals. This will use their SLAM as well as the predictive models that will be derived from it to assign waste collection and delivery routes as efficiently as possible. “We also use AIS [automatic identification systems] for additional collision avoidance with boats,” Deinema says. “The systems transmit their transponder information April/May 2020 | Unmanned Systems Technology A form of Lidar-based SLAM is used so that the USV can localise itself in GNSS-denied areas such as narrow spaces between tall buildings, or tunnels