UST030

7 Platform one UAV Navigation is developing autonomous sense & avoid technology for unmanned aircraft (writes Nick Flaherty). It is working on it with Sagetech Avionics and its MX family of transponders. The ground station will receive a specific warning when another platform is within a predefined limit or warn an operator with an alarm when traffic is really close. This is a concept of autonomous avoidance, said Miguel Angel de Frutos Carro, the company’s CTO. “We can sense the transponder of another craft, and the operator can manually change the flight plan to loiter and allow the other craft to pass by. This is challenging though with the current legislation – if we make a bad decision and enter a no-fly area, who is responsible? “We are therefore working with customers to develop the automatic decisions about changes in altitude, flight Engineers at MIT’s Computer Science and Artificial Intelligence Lab have developed a way to automatically design a controller for any hybrid UAV (writes Nick Flaherty). Hybrid UAVs have multiple modes of operation – vertical and horizontal flight, transitions and gliding – so the central controller has to operate differently in each mode but still keep the craft in the air. This can be a major challenge for system developers, as one control system directs the vehicle’s vertical rotors for hovering while another directs the horizontal rotors for speed and distance; this mode has very different flight dynamics. As a result, the controllers plans or an emergency loiter point if there is an alert,” he said. “For example, changing altitude is easier to control than lateral movement, so we implement a hover command for a rotary craft to give the operator more time, and for a fixed-wing craft we use loitering. We have a safety altitude that makes it easier for a manned aircraft to detect us.” have typically been designed manually – a time-consuming process. The team therefore looked at the challenge of automatically designing a single controller for the different flight modes, and how to generalise that method for any UAV model, shape or structure. “With this method, any non-expert could input their new UAV model into the system, wait a few hours to compute the controller, then have their own customised UAVs fly in the air,” said Jie Xu, one of the engineers. “It can make hybrid UAVs far more accessible to everyone.” The approach uses a controller based around a neural network that is trained by reinforcement learning techniques. The altitude depends on the platform. For example, manned aviation in Spain is not allowed under 150 m, so the first option is to go below that, or reach a specific safe altitude. That altitude has to be defined before executing the flight plan, as it is different in each region of Europe, such as near an airfield or over the sea. The system developer first decides on the geometry of a hybrid UAV by selecting and matching parts from a provided data set. The resulting design is then used in a simulator to automatically compute and test the UAV’s flight performance. A reinforcement learning algorithm is then applied to the controller to achieve the best performance in the simulation. The team has shown that this approach works for both simulations as well as real flight tests. In future work, the team intends to investigate how to increase manoeuvrability through improved geometry design – shape and positions of rotors/wings – to help improve the flight performance of the hybrid UAVs. Avoidance by transponders Auto-design for hybrid UAVs Airborne vehicles Airborne vehicles Unmanned Systems Technology | February/March 2020 Linking RF transponders and a UAV autopilot to make autonomous decisions is a key challenge (Courtesy of UAV Navigation)