Hazard-detection and avoidance systems will become an important asset for next-generation landing and exploration missions. To date, multiple studies into these systems were conducted to develop the methods needed and to demonstrate their performance in mainly software-based tests. Few studies were able to demonstrate the performance of the algorithm in hardware-in-the-loop tests, as these are usually difficult to set up and expensive to execute. In this paper, the hardware-in-the-loop testing of a stereo-vision based hazarddetection algorithm is presented. It was performed with the Testbed for Robotic Optical Navigation (TRON) at the German Aerospace Center (DLR) in Bremen, Germany. Since this testbed only allows for testing in a scaled environment, one of the challenging tasks was to design a scaled-down test set-up to represent a real-life lunar descent. The hardwarein- the-loop testing confirmed the results obtained during the earlier software-in-the-loop testing, that stereo vision can successfully be used for hazard detection during planetary descent.

Landing autonomously in hazardous environments is a very likely scenario for future exploration missions. Landing in hazardous but scientifically interesting sites on Mars or the Moon, returning to the surface of Venus or a landing on Europa are just a few examples for missions where hazards might be encountered during landing. These missions will need the ability to sense surface hazards, to select a safe landing site, and to avoid the detected hazards during touchdown. This will require autonomous landing site evaluation, but also more accurate navigation capabilities than the current state-of-the-art. Building upon our previous developments in the field of hazard detection and landing site evaluation, we developed a navigation filter capable of limiting the relative error with respect to detected features such as hazards and relative to selected points such as a safe landing site in the surface plane, as well as reducing any absolute navigation error accumulated in the altitude measurements. Using an error-state Kalman filter and measurements based on images and surface DEMs obtained from a hazard-detection method, we were able to greatly improve both landing accuracy and landing precision with respect to the current state of the art. With our filter we are able to reduce the hazard relative landing ellipse size by a factor of 3, while also reducing the ranging error to the surface by almost 99% thus enabling accurate altitude estimation during the descent. The developed method proofed to be robust with less than 1% of outliers created. Performing a hardware-in-the-loop test at the TRON facility at DLR Bremen concluded the work. The results of the test verified the results from the software-in-the-loop testing. This shows that hazard relative navigation techniques are a good candidate to enable a new class of exploration missions, capable of autonomous landing in unsafe and potentially even unknown landing regions.