Table of Content

Safe2Land's motivation and solution

A complete engine failure presents pilots with significant challenges. Under immense time and psychological pressure, two main problems in particular must be solved: First, a suitable landing site must be identified. If no official airport is within range, an appropriate emergency landing area must be selected. Second, the glide trajectory to that site must be calculated precisely, as it differs significantly from an approach with engine assistance.

With functioning engines, energy can be supplied at any time, while in a glide, only the available potential and kinetic energy of the aircraft, which continuously decreases, can be used. Therefore, it is necessary to model the aircraft's energy balance accurately so that the trajectory ends exactly at the runway threshold with the correct altitude and heading. Excess energy can be dissipated through specific flight maneuvers (e.g., turning) or the use of flaps. However, misjudgments of external conditions (e.g., wind or thermals) cannot be corrected by increasing thrust.

Currently available commercial landing assistance systems (such as Garmin, Foreflight, or SkyDemon) offer only limited support for these problems. Some systems (e.g., Garmin Autoland) still rely on engine power, while others only display the area that can be reached in a glide (e.g., Garmin SmartGlide or SkyDemon). However, an exact calculation of the glide trajectory does not take place, making these systems unsuitable for fully autonomous applications. Likewise, no known assistance system provides data on emergency landing sites outside registered airports – a crucial factor when altitude is too low for an extended glide.

The Safe2Land system we developed solves both of the above problems. By using pattern recognition techniques based on artificial neural networks, suitable emergency landing sites are identified from geodata (digital orthophotos, digital surface models, OpenStreetMap). For example, over 103,000 potential emergency landing sites have been identified in the state of North Rhine-Westphalia, which are stored in an emergency landing site database.

Based on our concept of kinematoide chains, glide trajectories are created that efficiently and precisely model environmental changes along the flight path, such as air density or altitude wind vectors. Safe2Land generates control instructions in the form of target heading and target bank, which are displayed to the pilot via a flight director. Additionally, we demonstrate that the system enables fully autonomous landings by translating the control instructions into inputs for an autopilot. This has been successfully tested in a flight simulator.

In the presentation, we will showcase the Safe2Land system using a workbench coupled with a flight simulator, allowing for quick evaluations of the trajectory algorithms. The parameters used for modeling the glide characteristics (e.g., optimal glide speed, glide ratio, wind direction, and strength at different altitudes) can only be estimated imprecisely and may change during the approach. Additionally, there are interactions: an overestimated glide ratio affects the glide performance similarly to a correct glide ratio combined with unexpected updrafts. With the workbench, we can examine how such disturbances can be compensated by dynamically adjusting the trajectory. For example, deploying the flaps earlier during the final approach can mitigate the effects of updrafts.

If these disturbances become too severe, a recalculation of the trajectory may be necessary. We recognize this need based on the predicted touchdown point relative to the (emergency) runway threshold, as determined by the kinematic chains. If, for example, the touchdown point is too far behind the threshold, the flaps are deployed earlier. Conversely, flap deployment is delayed or omitted if the predicted touchdown point no longer lies on the runway.

The Safe2Land system provides reliable support for pilots in the event of an engine failure and is also capable of performing the emergency landing fully autonomously via the autopilot.

Components of Safe2Land

Selected Presentations

Invited speech, IEEE BigComp Conference, Bangkok 2024

Emergency in an Austrian Alps' Valley

Deutscher Luft- und Raumfahrt-Kongress (DLRK 2024): 

"Dynamic adaptation and test bench evaluation of Engine-off emergency landing routes"

Speech summary: Today, we want to address a critical challenge in aviation: what happens when a complete engine failure occurs. This situation places pilots under enormous pressure, both in terms of time and mental strain. They must solve two critical problems almost immediately.

First, they need to identify a suitable landing site. If no official airport is within range, the pilot must choose an appropriate emergency landing area. Second, and equally important, is the precise calculation of the glide trajectory to reach that site—something that differs significantly from a standard, engine-assisted landing.

Under normal conditions, when the engines are functioning, energy can be added whenever needed. But during a glide, only the aircraft’s existing potential and kinetic energy are available—and they are constantly decreasing. This makes it essential to model the energy balance of the aircraft accurately. The trajectory must be designed to end at the exact runway threshold, with the correct altitude and heading. Excess energy can be dissipated through flight maneuvers like turning or deploying the landing flaps. However, any misjudgments—whether it’s wind or thermals—cannot be corrected by simply increasing power.

The problem is that most commercial landing assistance systems available today, such as Garmin, Foreflight, or SkyDemon, offer limited support. Some systems, like Garmin Autoland, still require engine power, while others only indicate the range you can reach in a glide, such as Garmin SmartGlide or SkyDemon. Unfortunately, none of these systems calculate the exact glide trajectory, making them unsuitable for fully autonomous applications. Additionally, no system currently provides data on emergency landing fields outside of registered airports—this is a crucial factor when altitude is too low for a long glide.

This is where Safe2Land, the system we developed, comes into play. It solves both of these issues. Using advanced pattern recognition techniques, based on artificial neural networks, Safe2Land identifies suitable emergency landing sites from geodata, such as digital orthophotos, digital surface models, and OpenStreetMap. To give you an example, in the state of North Rhine-Westphalia alone, we identified over 103,000 potential emergency landing sites, all stored in a specialized database.

But identifying the landing site is just part of the solution. Using our concept of kinematoide chains, Safe2Land also calculates glide trajectories that model environmental changes along the flight path, such as air density or wind vectors at different altitudes. The system generates precise control instructions—like target heading and bank angle—which are displayed to the pilot through a flight director. What’s more, Safe2Land enables fully autonomous landings by converting these instructions into commands for an autopilot. This capability has been successfully tested in a flight simulator.

During our presentation today, we will demonstrate the Safe2Land system using a workbench that is integrated with a flight simulator. This allows us to quickly evaluate the effectiveness of our trajectory algorithms. It’s important to note that while we use parameters such as optimal glide speed, glide ratio, and wind conditions at various altitudes to model the glide characteristics, these can only be estimated and may change during the flight. There are also interactions to consider. For example, an overestimated glide ratio can affect the glide path in much the same way as an unexpected updraft would. Using our workbench, we can analyze how such disturbances can be compensated for by dynamically adjusting the trajectory. In some cases, deploying the flaps earlier during the final approach can reduce the impact of an updraft.

However, if these disturbances become too strong, recalculating the trajectory might be necessary. We monitor this by comparing the predicted touchdown point to the threshold of the (emergency) runway. If the touchdown point is predicted to be too far beyond the threshold, Safe2Land will deploy the flaps earlier. Conversely, if the predicted touchdown point no longer falls on the runway, flap deployment is delayed or even skipped.

In conclusion, the Safe2Land system offers pilots a reliable tool for managing engine failure scenarios. Not only does it provide critical guidance, but it is also capable of performing the emergency landing entirely autonomously through the autopilot.