32 to be commercially viable, but Lidar prices have dropped dramatically, to the point that there’s really no excuse not to use one, given that one is enough for localisation. “If you analyse the increases in the various metrics for robustness that Lidar gives you, and measure those against the increase in the cost of the product, there’s no question that it’s worth it.” Tome has opted for Ouster Lidars since his days in the DARPA challenge, and its 500 g OSO-32 Wide Range Lidar is now typically mounted on the Keyper. It generates 1,310,720 points per second, at a maximum data rate of 86.55 Mbit/s, with a 90o vertical FoV (360o horizontal). Each point includes a measurement for range, reflectivity and azimuth angle among other parameters, along with a timestamp usually within 1 µs. “As well as Ouster building its product on many proprietary technologies, its Lidar has a really big FoV, scattering its points very far with each shot, and that’s crucial to triangulating our position at all times,” Tome says. First-time users can provide the Keybotic with a pre-made map of their industrial environments, which can be uploaded to the UGV as a localisation reference source. However, the company recommends that an on-site inspector guides the UGV around using a remote control console so that it can perform stereo camera and Lidar measurements of its surroundings, and use its computer vision to recognise and annotate critical objects such as gauges, valve wheels and so on for future reports. “That way, the Keyper gains way more information than many pre-existing digital twins,” Tome says. “To have as layered a virtual environment as possible, we use a multimodal sensor fusion system,” he adds. “If a sensor breaks down, the others act redundantly, and if something is hard to discern in one, chances are you can verify it with another. “It’s a complicated approach for sensor fusion, especially if you have different suppliers or teams for different sensors, but as I programmed the interfacing between everything myself since DARPA, and we have so many sensors complementing each another, it made complete sense for us.” Indoor autonomous robots are often limited to sparse maps with fixed navigational waypoints and highly automated behaviour, with little or no capacity for deviating from preprogrammed paths, even where prudent. Such is the detail of the Keybotic’s generated environment however that the resulting 3D map enables the Keyper to perform real-time autonomous waypoint changes and path repetitions or double takes, without operator interventions. “Of course, planning a path through a map with a billion cells is impossible, but just as Google Maps determines routes first by searching through each state, then by city, then by street, rather than looking at every single block and street in its database, our representation October/November 2023 | Uncrewed Systems Technology Intel RealSense cameras give the Keyper its perception for safe walking and balance, while an Ouster Lidar provides its localisation The Keyper’s battery bus is 48 V, which suits the servos and ESCs, while a power monitoring and distribution board ensures the sensors and computers get their respective voltages below 48 V
RkJQdWJsaXNoZXIy MjI2Mzk4