Accurate object detection is important for the safe navigation of autonomous vehicles (AVs) in dynamic environments. Conventional AVs rely on high-resolution but bulky, topmounted 3D Light Detection And Ranging (LiDAR) sensors, which are impractical for scalable production and computationally expensive to process. To address this, we utilize quasi-2D LiDAR sensors, which offer a compact and scalable solution for AVs. However, point clouds generated by these sensors have a narrow vertical field of view and contain a limited number of points compared to 3D LiDARs, which makes object detection significantly more difficult. To overcome this challenge, we transform the quasi-2D point clouds into Bird's-Eye-View (BEV) images and apply four different input encoding strategies. These encodings enhance the spatial and temporal context of the data, making them suitable for efficient deep-learning-based 2D object detection models to perform real-time detection with minimal computational overhead. In contrast to other approaches for 2D LiDARs, we are able to detect not only vehicles but also pedestrians and cyclists. We use a large real-world dataset to compare our proposed input encodings combined with three different state-of-the-art 2D object detection models. Our results demonstrate that these input encodings enhance object detection performance across all models, significantly reducing false positives, improving heading angle prediction, and boosting the detection accuracy of smaller traffic participants.

Control, tracking, and obstacle detection algorithms for mobile robots, including autonomous cars, rely on a jump-free estimate of the vehicle's pose. While one cannot completely avoid jumps in global solutions like INS/GNSS and SLAM, relative localization (i.e., odometry) does not suffer from this problem. Methods based on graph optimization are popular in that field, but they do not scale very well with high-frequency measurements. Kalman filters (KFs) are able to cope with those measurements, but they face the issue of a continuously growing covariance. This results in instabilities and eventually jumps in the state estimate. We present an approach to handle this problem by periodically moving the reference state forward in time, which is realized using two filters. The equations for implementing this in both the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are derived. The algorithm is evaluated using real-world datasets covering different scenarios of autonomous driving. We show that our method provides a smooth and stable estimate even over long time periods and that it achieves a better localization performance than the standard approach.

With the rise of advanced driver assistance systems and highly automated driving, the research of highly reliable and accurate tracking algorithms received special attention within the last years. While most of the investigation treats the objects to be tracked as single points, the appropriate handling of uncertainties, when also the object's extent is of interest, has been ignored so far. Classical approaches do not capture the properties of the probability density function, which in fact resembles a banana-shaped distribution, when handling the object's state and the measurements in the Euclidean space. In this paper, we tackle this problem, called extended object tracking, by tracking the object's state on the affine group which belongs to the matrix Lie groups. By incorporating the second-order dynamics of the movement, we model the state space as the direct product Aff(2) ×\mathbbR3and handle the uncertainties on the associated Lie algebra to establish a proper representation. We derive the necessary equations to use the extended Kalman filter on Lie groups including an appropriately modeled measurement space. We show the advantage of our proposed method by comparing it to a classical approach, which treats both the state and the measurement space in the Euclidean space. The results of the simulations prove that, by encapsulating the state and measurement space in the appropriate manifold, we obtain higher stability and realibility with smaller state estimation errors.

Keeping track of the current state is a crucial task for mobile autonomous systems, which is referred to as state estimation. To solve that task, information from all available sensors needs to be fused, which includes relative measurements as well as observations of the surroundings. In a dynamic 3D environment, the pose of an agent has to be chosen such that the most relevant information can be observed. We propose an approach for multi-sensor fusion and active perception within an autonomous deep space navigation scenario. The probabilistic modeling of observables and sensors for that particular domain is described. For state estimation, we present an Extended Kalman Filter, an Unscented Kalman Filter, and a Particle Filter, which all operate on a manifold state space. Additionally, an approach for active perception is proposed, which selects the desired attitude of the spacecraft based on the knowledge about the dynamics of celestial objects, the kind of information they provide as well as the current uncertainty of the filters. We evaluated the localization performance of the algorithms within a simulation environment. The filters are compared to each other and we show that our active perception strategy outperforms two other information intake approaches.