IDV Viking UGV | Digest (MACE), including the main autonomy computer, which IDV-R has designed and engineered in-house to meet military standards on ruggedness and power conditioning. MACE can be configured in various form factors for each vehicle integration, but is likely to be standardised in future with a two-box configuration, with one computer executing basic functions and the other carrying out more complex, autonomy related processing. Passive GNSS- denied navigation Each of the sensors discussed also plays a part in IDV-R’s ATLAS system for GNSS-denied navigation. The frequent problem of trees and buildings near the UGV hampering its satellite reception from GNSS constellations, combined with the ever-increasing threat of electronic warfare, jamming and spoofing, motivated the creation of ATLAS. “We decided in 2016 that we ought to try using computer vision on the UGV to observe landmarks and compare them with an onboard map database to get our solution,” Maloney says. “We started using a probabilistic filter to try and match up landmark observations with the map database to figure out where the UGV was, to navigate just like a soldier would.” The team won funding from the UK Defence and Security Accelerator to pursue a proof-of-concept, and it so happened that the first generations of NVidia Jetson embedded-AI computers were being released. These provided the ideal computing power and architecture for training Viking to recognise landmarks using a neural network. ATLAS has since progressed through several phases and trials with the UK MoD and British Army. Last year, it was tested on the Cultana range in the Australian outback, with Viking under real GNSS jamming conditions as part of the Aukus Trusted Operation of Robotic Vehicles in Contested Environments programme. “For training the neural networks, we do image gathering and annotation twice: the first round is done from above, using aerial or satellite systems to collect imagery, in which we label anything that could be used as a landmark and train the networks. Once the aerial network is trained, we can run it across new aerial imagery, creating a map database that Viking can leverage. We also vectorise it and make it semantic, so it’s small and efficient for fitting onboard the vehicle’s data storage,” Maloney says. IDV-R also supplies map tools, so that when customers go somewhere new, they can source their own imagery and create the map databases for ATLAS and strategic route planning. “Then, the UGV observes landmarks in the field, and uses AI to classify anything it sees that qualifies as a landmark; for example, the edge of a track, a road, tree or building. It can then compare the range and bearing to those with the map database’s spatial information. With a good few landmarks around, the system can localise itself with accuracy approaching 10 cm, which is good enough for autonomous guidance,” Maloney adds. During movement, the UGV continuously performs visual measurements with the stereo camera (and Lidar if the situation allows), using that data to update the particles of its probabilistic filter and localise itself, combining that localisation analysis with attitude and heading information from its IMU and odometry, resulting in a probability distribution of where the UGV thinks it is. “We also use AI for real-time terrain perception, looking ahead and recognising optimal routes for getting between waypoints; for instance, having the AI decide where it’s best to drive, and avoiding obstacles within constraints around a planned route from the user or global/strategic planner,” Maloney says. “The mission itself, global or strategic route, is usually planned by an operator using our Base Station software. The map database we use can currently be used to generate routes to an objective. The operator can set waypoints and a range of preferences, and the algorithms will work out a route.” The map data and control software are installed on standard SSDs for secure storage against shock and vibration, with some work having gone into cybersecurity and remote purging for military users who want to keep their tech out of hostile hands. 57 Uncrewed Systems Technology | June/July 2024 IDV-R’s modular autonomous control equipment (MACE) includes the main autonomy computer, designed to meet military standards on ruggedness and power conditioning
RkJQdWJsaXNoZXIy MjI2Mzk4