Unmanned Systems Technology 024 | Wingcopter 178 l 5G focus l UUVs insight l CES report l Stromkind KAT l Intelligent Energy fuel cell l Earthsense TerraSentia l Connectors focus l Advanced Engineering report

79 EarthSense TerraSentia | In operation users for the system to use Ackermann steering (which is similar to that on tractors and other farm equipment) for the sake of mechanical commonality. Simplicity, however, ended up taking precedence, hence the use of skid- steering. The TerraSentia drives through each row at about 1.6 kph, or 1 m every 2.27 s. While the operating speed could double or triple over the next couple of years as development proceeds, that depends on the ability of the camera and embedded software to update faster. “If you’re looking for really slight changes in the pixels, such as spots, minor discolouration or disease markers, then you’d probably keep to slow speeds, but if you’re just looking for plant height you could go much faster,” Chowdhary says. While the current wheels use a paddle- like design for traction, the team is replacing them with a soft, resin-based design. These can deform under load but are sufficiently rigid and stiff to withstand vertical stresses and bending loads, the latter being critical to be able to skid-turn the wheels independently on the ground. Notably, unlike UAVs, the TerraSentia might be better suited to operating at night than in the day. Lidar emits its own light for measurement and navigation, while GNSS operates independently of the Sun. And two illuminators on the vehicle provide enough light for the vision cameras, without the shadows and other sunlight-related issues that can occur in the day. “Some types of data collection are actually better done at night,” Chowdhary says. “Researchers who collect carbon data like to do it at night, because that’s when plants transpire.” Post-operation Once the fields have been successfully measured (or a battery needs replacing), the TerraSentia autonomously returns to a maintenance barn, which the team is developing as an automated battery swap station. In addition to hot-swapping a charged battery, the barn will spray- clean the wheels and sensors. Data is collected via an Ethernet port inside the battery compartment. A cable can then be run from this to a computer with an internet connection, and the data will be uploaded to the EarthSense cloud server. With a typical bandwidth, a day’s worth of data – 6-7 Gbytes – can be uploaded in a few hours. Users can then access their data through their account on the aforementioned app, with the data firewalled so that no-one else can access it. The company’s analytics software can then automate the analysis of the phenotypic variables being studied, and output it into a GeoJSON file with the geo-tagged data in an actionable format. “A big impetus for us is on fusing all the different data streams on the TerraSentia, to improve the analytics. Our initial analytics were on corn count – where we were getting accuracy past 80%, sometimes 90% depending on certain factors – but with our new methods, where we factor in the vehicle’s speed and keep tracking the corn through frames, we can get 95% accuracy on young corn,” Chowdhary notes. “Height is already very accurate because of the precise nature of Lidar measurements. Stem width is pretty challenging, especially with cameras because of occlusion, but accuracy is improving as we find other ways of removing the background, using the speed of the vehicle and motion in the frame as indicators of what’s there and what’s not.” The future Farming work tends to be seasonal, so the next phase of early adopter trials of the TerraSentia will begin around this April, and run until September. “Before that though we’ll be optimising the design, going over the data from last year, improving our analytics algorithms and getting ready to engage with our customers,” Chowdhary says. “We are working with the University of Illinois at Urbana-Champaign, Cornell University and UW Madison in Colorado, and we have two major breeding firms on board, of which there are perhaps only four in the world.” With enthusiastic support and practical development set to continue, the TerraSentia looks poised to cultivate some major changes in agriculture. Unmanned Systems Technology | February/March 2019 The current wheel design on TerraSentia uses a paddle-like tread for providing traction over soft earth, but that is being phased out in favour of a soft resin wheel design for improved skid-turning