Robust and accurate visual-inertial estimation is crucial to many of today's challenges in robotics. Being able to localize against a prior map and obtain accurate and drift-free pose estimates can push the applicability of such systems even further. Most of the currently available solutions, however, either focus on a single session use case, lack localization capabilities, or do not provide an end-to-end pipeline. We believe that only a complete system, combining state-of-the-art algorithms, scalable multisession mapping tools, and a flexible user interface, can become an efficient research platform. We, therefore, present maplab, an open, research-oriented visual-inertial mapping framework for processing and manipulating multisession maps, written in C++. On the one hand, maplab can be seen as a ready-to-use visual-inertial mapping and localization system. On the other hand, maplab provides the research community with a collection of multisession mapping tools that include map merging, visual-inertial batch optimization, and loop closure. Furthermore, it includes an online frontend that can create visual-inertial maps and also track a global drift-free pose within a localization map. In this letter, we present the system architecture, five use cases, and evaluations of the system on public datasets. The source code of maplab is freely available for the benefit of the robotics research community.

In the context of future urban automated driving many important problems remain unsolved. A critical one is the analysis and prediction of pedestrian movements around urban roads. Especially the analysis of non-critical situations has not received much attention in the past. This paper focuses on analyzing and predicting movements of pedestrians approaching crosswalks, a very crucial pedestrian-vehicle interaction in urban scenarios. In our previous work, we analyzed the performance of a data-driven Support Vector Machine-based architecture, and the relevance of specific features to infer pedestrian crossing intentions. In this paper, we will use our previous results as baseline to compare against an architecture based on neural networks for time-series classification. In particular we analyze the effectiveness of dense and Long-Short-Term-Memory networks. Furthermore, we will be looking into enhancing our feature vectors by adding LiDAR based images to the classification process. Additionally the evaluation provides an estimate for the temporal prediction horizon. The approaches presented are validated with real world trajectories recorded in Germany. Our results show an average accuracy improvement of 10-20% with respect to our previous Support Vector Machine-based approach.

In this work, we propose a novel approach for integrating rules into traffic agent trajectory prediction. Consideration of rules is important for understanding how people behave-yet, it cannot be assumed that rules are always followed. To address this challenge, we evaluate different approaches of integrating rules as inductive biases into deep learning-based prediction models. We propose a framework based on generative adversarial networks that uses tools from formal methods, namely signal temporal logic and syntax trees. This allows us to leverage information on rule obedience as features in neural networks and improves prediction accuracy without biasing towards lawful behavior. We evaluate our method on a real-world driving dataset and show improvement in performance over off-the-shelf predictors.

Most autonomous driving solutions require some method of localization within their environment. Typically, onboard sensors are used to localize the vehicle precisely in a previously recorded map. However, these solutions are sensitive to ambient lighting conditions such as darkness and inclement weather. Additionally, the maps can become outdated in a rapidly changing environment and require continuous updating. While LiDAR systems don't require visible light, they are sensitive to weather such as fog or snow, which can interfere with localization. In this paper, we utilize a Ground Penetrating Radar (GPR) to obtain precise vehicle localization. By mapping and localizing using features beneath the ground, we obtain features that are both stable over time, and maintain their appearance during changing ambient weather and lighting conditions. We incorporate this solution into a full-scale autonomous vehicle and evaluate the performance on over 17 km of testing data in a variety of challenging weather conditions. We find that this novel sensing modality is capable of providing precise localization for autonomous navigation without using cameras or LiDAR sensors.