Unmanned Systems Technology 022 | XOcean XO-450 l Radar systems l Space vehicles insight l Small Robot l BMPower FCPS l Prismatic HALE UAV l InterDrone 2018 show report l UpVision l Navigation systems

52 Digest | The Small Robot Company it travels around the fields every couple of weeks. All the data (of which there are gigabytes) is stored on a USB stick and then transferred to a system in the kennel. Tom is controlled by the open source Robot Operating System (ROS), where modules are available for a range of motor control and image capture functions, and runs on a Raspberry Pi single-board computer. The ROS also collects data from encoders on the wheels that provide 7000 data points, or ‘ticks’, per revolution, and this is combined with data from an onboard accelerometer to provide inertial navigation alongside the GNSS RTK data. Small Robot is working with other single-board computers that use the Arduino form factor, says Joe Allnut, chief roboticist at Small Robot. Arduino has different connectors to the Raspberry Pi, allowing Tom to use other types of encoders in the Arduino format. The company has also developed an operating system called Wilma, which is based on Ubuntu Linux and runs on a second Raspberry Pi on Tom as the main controller. Small Robot is also looking at the variant of Ubuntu that is aimed at systems for the Internet of Things. This would add more integrated networking and better control of sensors. Tom is powered by nickel metal hydride batteries as these are more reliable for long-term use and allow unmonitored charging in the kennel. At the moment, the batteries are charged in the kennel and replaced manually in Tom when necessary. A wireless charging system would allow Tom to charge itself but it would require accurate positioning within the kennel and more charging technology on the platform, increasing the weight. Instead the company has opted for a manual changeover of batteries to simplify the hardware design. Back in the kennel, the data on the USB stick is transferred to a standard PC with a GPU accelerator running Windows. This runs a machine learning (ML) algorithm called NOUS that has been developed with Cosmonio in Cambridge. The algorithm identifies the plants and their state of health, learning as the year goes on to improve the accuracy of the detection of problems in particular fields and uploading the results to the cloud to make use of the learning from all the other farms that Small Robot supplies. Machine learning Systems are currently out in fields at 20 farms collecting data for the ML system to develop a digital crop model that learns the best time to send out Tom and the optimum time to call for help from the larger systems, whether it’s for fertilising or weeding. This data is also used to optimise the type of fertiliser needed for the crops, according to how well they are growing. These larger systems are all based on a chassis called Jack, which is powered by lithium-ion batteries. The design of the chassis faces particular challenges. It has to be small enough to fit into a Transit van to be delivered to a farm when it is needed, but it also has to take a standard 3 m boom for the seeding and fertiliser delivery, and be light enough to have minimal impact on the ground as it moves around. The design therefore has a series of articulated arms to enable it to be folded up into the back of a van, then unfolded to take the boom. The wheels have a diameter of 21 in, or 54 cm ±2 cm, to minimise the pressure on the October/November 2018 | Unmanned Systems Technology A number of different structures have been evaluated for the autonomous farming systems. This variant, Dick, is for feeding and watering crops Systems are in fields collecting data to develop a digital crop model and to optimise the type of fertiliser the crops need