UGVs | Insight offered as solutions. They’re coming, definitely, but you need to start with an autonomous implement for going behind a tractor, crewed or uncrewed.” Newman and AROW can both weed multiple rows of a field at once, with the typical configuration using one 3D stereo camera, one RGB camera and one pair of implements per row. The standard version features six pairs of implements, spaced equally along a horizontal length. As the platform pulls Newman through a field, AROW’s cameras recognise the crops, having been trained through image-based artificial neural networks, and they send commands to Newman’s implements to prune any plants that are not known. Ullmanna also performs its own image gathering for AI training, taking around 1000 photos per hour for a new crop species (although the company continually takes more images to improve its crop-recognition models). It manually annotates them to say which plants are that crop and which are not, before feeding them into training software for crop classification. “Once AROW sees a cluster of plants, based on the movement speed of the vehicle and the locations of the noncrops, it sends commands to the arms mounting the knives and then to the solenoids, which control their opening and closing motions for cutting the weeds,” Ullmann says. “It is very complex: the knives have to cut as close as possible to the crop centres without damaging them to be able to remove every last weed. We’ve had to integrate a lot of premium-price electronics to execute the cultivation, but one six-row Newman machine can provide the same cultivation throughput as 60 manual workers. “That’s not just weeding, but actual cultivation. AROW commands the knives to loosen the soil around the crops and that stops further weeds germinating around them.” He adds that one potential end-user is interested in using Ullmanna’s technology for detecting and recognising broccoli for harvesting. Each individual broccoli potentially grows at a different size and elevation, and tractors’ cutting implements must adjust for them, but AI training could achieve this. “With an AI-powered camera, like in our system, we can specify where an implement needs to aim and cut a broccoli based on its size and height. It’s just a matter of using 1000 images with different shapes and sizes of broccoli to get started,” Ullmann adds. Special qualities Across these various UGVs and their specialist uses, one sees some common themes behind how they are being designed to suit industry’s demands. Naturally, substituting humans in dull, dirty and dangerous work is common across many uncrewed systems applications, but in addition to these factors, some others stand out. One is ease of maintenance. Fast vanishing are the days when a team of 10 engineers in-the-loop and 12 specialist mechanics kept a self-driving truck or bus going. Longer endurance, more uptime and less cumbersome tasks are prized targets in today’s UGV developments. Another is comfort for humans, both inside and outside the autonomous vehicle. If UGVs are to become part of daily life, they must cause so little disruption and discomfort (whether stepping in and out of a robo-taxi or being wheeled around by a self-driving micro-mobility platform) that people barely notice the autonomous aspect of the system. Will other uncrewed vehicle types be able to match UGVs for these qualities? Making delivery multirotors quieter, or helicopters or boats less maintenance-intensive are challenging targets to hit, albeit worthwhile ones for aerial or aquatic mobility. But, if it pays to specialise and convenience is king, one can expect UGVs to continue being innovated for highly specialised and valuable new applications in the years ahead. 67 Uncrewed Systems Technology | April/May 2024 Ullmanna’s Newman cultivator and AROW sensor box are trained to recognise and intelligently weed around crops with the throughput of up to 60 seasonal workers (Image courtesy of Ullmanna)
RkJQdWJsaXNoZXIy MjI2Mzk4