USE Network launch I UAV Works VALAQ l Cable harnesses l USVs insight l Xponential 2020 update l MARIN AUV l Suter Industries TOA 288 l Vitirover l AI systems l Vtrus ABI

88 Focus | AI systems combined with data from a weather satellite and a rule management system called the Operational Decision Manager (ODM) to follow the International Regulations for Preventing Collisions at Sea (COLREGs) as well as recommendations from the International Convention for the Safety of Life at Sea. A real-world scenario used for training is to have the Mayflower in the open ocean, approaching the US coast at Cape Cod, with no satellite connectivity. In its path ahead is a cargo ship that has collided with a fishing vessel and spilt some of its load. In this scenario, the AI system assesses the current environment and uses radar to identify hazards at a range of 2.5 nautical miles. Onboard cameras feed the neural network that identifies hazards such as a cargo ship, a fishing vessel and three partially submerged shipping containers floating in the water. An Automatic Identification System provides information about the cargo ship’s class, weight, speed and cargo, and links that with data from a GNSS navigation system providing current location, heading, speed and course. Attitude sensors on the craft assess the local sea state and how the Mayflower pitches and rolls in the waves, while a vehicle management system provides operational data such as the battery charge level, power consumption and the power needs of the comms and science payloads. The ODM evaluates the COLREGs with respect to the other vessels in the vicinity and generates a risk map indicating an ‘unsafe’ situation ahead; this is combined by a piece of software called the AI Captain with all the other data. A safety manager function reviews all the AI Captain’s decisions to ensure they are safe for the Mayflower and other vessels in its vicinity. The resulting decision, for example to steer to starboard to avoid the unexpected navigation hazard, is verified by the safety manager, and the AI Captain then instructs the vehicle management system to change course and speed. Aerial One of the new neural network architectures, SNN, is applicable to UAV systems. It has been used for a low- power anti-collision radar system for UAVs to identify approaching objects in a matter of milliseconds. This spiking network mimics the way groups of biological neurons operate to recognise temporal patterns – the neuromorphic approach referred to above. This can result in a power consumption 100 times lower than traditional implementations with a tenfold reduction in latency. The approach can identify micro- Doppler radar signatures using only 30 mW of power, while the chip’s architecture and algorithms can easily be tuned to process a variety of sensor data including sonar, radar and Lidar streams. SNNs operate very similarly to biological neural networks, in which neurons fire electrical pulses sparsely over time and only when the sensory input changes. As such, energy consumption can be reduced significantly. What’s more, the spiking neurons can be constantly reconfigured, turning the SNN into a dynamic system that learns and remembers temporal patterns. The event-driven digital design makes the chip behave exactly and repeatedly as predicted by the neural network simulation tools. In one scenario being developed, UAVs are using their onboard camera and radar sensor systems for in-warehouse navigation, keeping a safe distance from walls and shelves while performing complex tasks. The technology could be used for other applications, such as deploying AGVs. June/July 2020 | Unmanned Systems Technology A prototype chip called Loihi implements a spiking neural network architecture (Courtesy of Intel)