This paper describes a novel approach to simultaneous localization and mapping (SLAM) techniques applied to the autonomous planetary rover exploration scenario to reduce both the relative and absolute localization errors, using two well‐proven techniques: particle filters and scan matching. Continuous relative localization is improved by matching high‐resolution sensor scans to the online created local map. Additionally, to avoid issues with drifting localization, absolute localization is globally corrected at discrete times, according to predefined event criteria, by matching the current local map to the orbiter's global map. The resolutions of local and global maps can be appropriately chosen for computation and accuracy purposes. Further, the online generated local map, of the form of a structured elevation grid map, can also be used to evaluate the traversability of the surrounding environment and allow for continuous navigation. The objective of this study is to support long‐range low‐supervision planetary exploration. The implemented SLAM technique has been validated with a data set acquired during a field test campaign performed at the Teide Volcano on the island of Tenerife, representative of a Mars/Moon exploration scenario.

Rovers operating on Mars require more and more autonomous features to fulfill their challenging mission requirements. However, the inherent constraints of space systems render the implementation of complex algorithms an expensive and difficult task. In this paper, we propose an architecture for autonomous navigation. Efficient implementations of autonomous features are built on top of the ExoMars path following navigation approach to enhance the safety and traversing capabilities of the rover. These features allow the rover to detect and avoid hazards and perform significantly longer traverses planned by operators on the ground. The efficient navigation approach has been implemented and tested during field test campaigns on a planetary analogue terrain. The experiments evaluated the proposed architecture by autonomously completing several traverses of variable lengths while avoiding hazards. The approach relies only on the optical Localization Cameras stereo bench, a sensor that is found in all current rovers, and potentially allows for computationally inexpensive long‐range autonomous navigation in terrains of medium difficulty.

Autonomy on rovers is desirable in order to extend the traversed distance, and therefore, maximize the number of places visited during the mission. It depends heavily on the information that is available for the traversed surface on other planet. This information may come from the vehicle's sensors as well as from orbital images. Besides, future exploration missions may consider the use of reconfigurable rovers, which are able to execute multiple locomotion modes to better adapt to different terrains. With these considerations, a path planning algorithm based on the Fast Marching Method is proposed. Environment information coming from different sources is used on a grid formed by two layers. First, the Global Layer with a low resolution, but high extension is used to compute the overall path connecting the rover and the desired goal, using a cost function that takes advantage of the rover locomotion modes. Second, the Local Layer with higher resolution but limited distance is used where the path is dynamically repaired because of obstacle presence. Finally, we show simulation and field test results based on several reconfigurable and non-reconfigurable rover prototypes and a experimental terrain.

Accurate and complete terrain maps enhance the awareness of autonomous robots and enable safe and optimal path planning. Rocks and topography often create occlusions and lead to missing elevation information in the Digital Elevation Map (DEM). Currently, these occluded areas are either fully avoided during motion planning or the missing values in the elevation map are filled-in using traditional interpolation, diffusion or patch-matching techniques. These methods cannot leverage the high-level terrain characteristics and the geometric constraints of line of sight we humans use intuitively to predict occluded areas. We introduce a self-supervised learning approach capable of training on real-world data without a need for ground-truth information to reconstruct the occluded areas in the DEMs. We accomplish this by adding artificial occlusion to the incomplete elevation maps constructed on a real robot by performing ray casting. We first evaluate a supervised learning approach on synthetic data for which we have the full ground-truth available and subsequently move to several real-world datasets. These real-world datasets were recorded during exploration of both structured and unstructured terrain with a legged robot, and additionally in a planetary scenario on Lunar analogue terrain. We state a significant improvement compared to the baseline methods both on synthetic terrain and for the real-world datasets. Our neural network is able to run in real-time on both CPU and GPU with suitable sampling rates for autonomous ground robots. We motivate the applicability of reconstructing occlusion in elevation maps with preliminary motion planning experiments.

The exploitation of solar power for energy supply is of increasing importance. While technical development mainly takes place in the engineering disciplines, computer science offers adequate techniques for optimization. This work addresses the problem of finding an optimal heliostat field arrangement for a solar tower power plant. We propose a solution to this global, non-convex optimization problem by using an evolutionary algorithm. We show that the convergence rate of a conventional evolutionary algorithm is too slow, such that modifications of the recombination and mutation need to be tailored to the problem. This is achieved with a new genotype representation of the individuals. Experimental results show the applicability of our approach.