Safe2Land

Safe2Land helps in case of an engine-out or pilot's incapacitation. By means of the Emergency Landing Assistance (ELA) functionality we compute trajectories to close landing opportunities suited for the used aircraft type. Our new moving target approach just can compute 1 million (!) trajectories in one second one a COTS-tablet PC. Compared to other conventional methods our algorithm is more the 1000 times faster. Thus we can provide permanent recalculations of the trajectories in case that the wind changes.

Another component of Safe2Land is the Emergency Landing Field Identification (ELFI). Combining geodata like satellite images, LiDAR elevation data and street maps we can create databases of emergency landing fields if no public airfield is within reach. This is especially useful if a severe damage happened to the aircraft's structure (or fire) and an immediate landing is required. Here we mainly use artificial neural networks for improved accuracy even if the resolution of the elevation data is just in the range of 10 times 10 meter. Especially, we use topology-optimized artificial neural networks to find appropriate emergency landing sites. Recently, we have finished a database for two german federal states (e.g. North Rhine Westfalia with more than 102.000 emergency landing fields from 300-500 m length for small aircrafts).

Our current research for support of flight control (FC) focuses on a new kind of avionic instrument that we call Smart Flight Director (SFD). It was primarily developed to guide a pilot (or even a laymen) to safely follow the computed emergency trajectory. For SPO-cockpits it could be used to enable flight attendants to take over the control in case of a pilot's incapacitation. It could be used to allow laymen flying an airtaxi. And last but not least, SFD can also be facilitate and improve flight training. 

Beside SFD we are currently working towards artificial neural network pilots (ANNPs) thats are trained by reinforcement learning. At the last DLRK we had a session "Künstliche Neuronale Netze in der Luftfahrt (“Artificial Neural Networks in Aviation) and presented first promising results. By means of reinforcement learning we have trained artificial neural networks that find the best configuration (e.g. flaps setting and even slipping by crossed aileron and rudder) to bring the aircraft in gliding closely to the threshold. Moreover we also managed to train neural pilots that learnt to fly the aircraft by trial and error (of cause in simulation!).