Search

Automotive traffic scenes are complex due to the variety of possible scenarios, objects, and weather conditions that need to be handled. In contrast to more constrained environments, such as automated underground trains, automotive perception systems cannot be tailored to a narrow field of specific tasks but must handle an ever-changing environment with unforeseen events. As currently no single sensor is able to reliably perceive all relevant activity in the surroundings, sensor data fusion is applied to perceive as much information as possible. Data fusion of different sensors and sensor modalities on a low abstraction level enables the compensation of sensor weaknesses and misdetections among the sensors before the information-rich sensor data are compressed and thereby information is lost after a sensor-individual object detection. This paper develops a low-level sensor fusion network for 3D object detection, which fuses lidar, camera, and radar data. The fusion network is trained and evaluated on the nuScenes data set. On the test set, fusion of radar data increases the resulting AP (Average Precision) detection score by about 5.1% in comparison to the baseline lidar network. The radar sensor fusion proves especially beneficial in inclement conditions such as rain and night scenes. Fusing additional camera data contributes positively only in conjunction with the radar fusion, which shows that interdependencies of the sensors are important for the detection result. Additionally, the paper proposes a novel loss to handle the discontinuity of a simple yaw representation for object detection. Our updated loss increases the detection and orientation estimation performance for all sensor input configurations. The code for this research has been made available on GitHub. [ABSTRACT FROM AUTHOR]

State-of-the-art 3D object detection for autonomous driving is achieved by processing lidar sensor data with deep-learning methods. However, the detection quality of the state of the art is still far from enabling safe driving in all conditions. Additional sensor modalities need to be used to increase the confidence and robustness of the overall detection result. Researchers have recently explored radar data as an additional input source for universal 3D object detection. This paper proposes artificial neural network architectures to segment sparse radar point cloud data. Segmentation is an intermediate step towards radar object detection as a complementary concept to lidar object detection. Conceptually, we adapt Kernel Point Convolution (KPConv) layers for radar data. Additionally, we introduce a long short-term memory (LSTM) variant based on KPConv layers to make use of the information content in the time dimension of radar data. This is motivated by classical radar processing, where tracking of features over time is imperative to generate confident object proposals. We benchmark several variants of the network on the public nuScenes data set against a state-of-the-art pointnet-based approach. The performance of the networks is limited by the quality of the publicly available data. The radar data and radar-label quality is of great importance to the training and evaluation of machine learning models. Therefore, the advantages and disadvantages of the available data set, regarding its radar data, are discussed in detail. The need for a radar-focused data set for object detection is expressed. We assume that higher segmentation scores should be achievable with better-quality data for all models compared, and differences between the models should manifest more clearly. To facilitate research with additional radar data, the modular code for this research will be made available to the public. [ABSTRACT FROM AUTHOR]

In circuit motorsport, race strategy helps to finish the race in the best possible position by optimally determining the pit stops. Depending on the racing series, pit stops are needed to replace worn-out tires, refuel the car, change drivers, or repair the car. Assuming a race without opponents and considering only tire degradation, the optimal race strategy can be determined by solving a quadratic optimization problem, as shown in the paper. In high-class motorsport, however, this simplified approach is not sufficient. There, comprehensive race simulations are used to evaluate the outcome of different strategic options. The published race simulations require the user to specify the expected strategies of all race participants manually. In such simulations, it is therefore desirable to automate the strategy decisions, for better handling and greater realism. It is against this background that we present a virtual strategy engineer (VSE) based on two artificial neural networks. Since our research is focused on the Formula 1 racing series, the VSE decides whether a driver should make a pit stop and which tire compound to fit. Its training is based on timing data of the six seasons from 2014 to 2019. The results show that the VSE makes reasonable decisions and reacts to the particular race situation. The integration of the VSE into a race simulation is presented, and the effects are analyzed in an example race. [ABSTRACT FROM AUTHOR]

Applying an optimal race strategy is a decisive factor in achieving the best possible result in a motorsport race. This mainly implies timing the pit stops perfectly and choosing the optimal tire compounds. Strategy engineers use race simulations to assess the effects of different strategic decisions (e.g., early vs. late pit stop) on the race result before and during a race. However, in reality, races rarely run as planned and are often decided by random events, for example, accidents that cause safety car phases. Besides, the course of a race is affected by many smaller probabilistic influences, for example, variability in the lap times. Consequently, these events and influences should be modeled within the race simulation if real races are to be simulated, and a robust race strategy is to be determined. Therefore, this paper presents how state of the art and new approaches can be combined to modeling the most important probabilistic influences on motorsport races—accidents and failures, full course yellow and safety car phases, the drivers' starting performance, and variability in lap times and pit stop durations. The modeling is done using customized probability distributions as well as a novel "ghost" car approach, which allows the realistic consideration of the effect of safety cars within the race simulation. The interaction of all influences is evaluated based on the Monte Carlo method. The results demonstrate the validity of the models and show how Monte Carlo simulation enables assessing the robustness of race strategies. Knowing the robustness improves the basis for a reasonable determination of race strategies by strategy engineers. [ABSTRACT FROM AUTHOR]

Since 2017, a research team from the Technical University of Munich has developed a software stack for autonomous driving. The software was used to participate in the Roborace Season Alpha Championship. The championship aims to achieve autonomous race cars competing with different software stacks against each other. In May 2019, during a software test in Modena, Italy, the greatest danger in autonomous driving became reality: A minor change in environmental influences led an extensively tested software to crash into a barrier at speed. Crashes with autonomous vehicles have happened before but a detailed explanation of why software failed and what part of the software was not working correctly is missing in research articles. In this paper we present a general method that can be used to display an autonomous vehicle disengagement to explain in detail what happened. This method is then used to display and explain the crash from Modena. Firstly a brief introduction into the modular software stack that was used in the Modena event, consisting of three individual parts—perception, planning, and control—is given. Furthermore, the circumstances causing the crash are elaborated in detail.By presented and explaining in detail which software part failed and contributed to the crash we can discuss further software improvements. As a result, we present necessary functions that need to be integrated in an autonomous driving software stack to prevent such a vehicle behavior causing a fatal crash. In addition we suggest an enhancement of the current disengagement reports for autonomous driving regarding a detailed explanation of the software part that was causing the disengagement. In the outlook of this paper we present two additional software functions for assessing the tire and control performance of the vehicle to enhance the autonomous. [ABSTRACT FROM AUTHOR]

Connected and autonomous vehicles (CAVs) could reduce emissions, increase road safety, and enhance ride comfort. Multiple CAVs can form a CAV platoon with a close inter-vehicle distance, which can further improve energy efficiency, save space, and reduce travel time. To date, there have been few detailed studies of self-driving algorithms for CAV platoons in urban areas. In this paper, we therefore propose a self-driving architecture combining the sensing, planning, and control for CAV platoons in an end-to-end fashion. Our multi-task model can switch between two tasks to drive either the leading or following vehicle in the platoon. The architecture is based on an end-to-end deep learning approach and predicts the control commands, i.e., steering and throttle/brake, with a single neural network. The inputs for this network are images from a front-facing camera, enhanced by information transmitted via vehicle-to-vehicle (V2V) communication. The model is trained with data captured in a simulated urban environment with dynamic traffic. We compare our approach with different concepts used in the state-of-the-art end-to-end self-driving research, such as the implementation of recurrent neural networks or transfer learning. Experiments in the simulation were conducted to test the model in different urban environments. A CAV platoon consisting of two vehicles, each controlled by an instance of the network, completed on average 67% of the predefined point-to-point routes in the training environment and 40% in a never-seen-before environment. Using V2V communication, our approach eliminates casual confusion for the following vehicle, which is a known limitation of end-to-end self-driving. [ABSTRACT FROM AUTHOR]

Typically, lane departure warning systems rely on lane lines being present on the road. However, in many scenarios, e.g., secondary roads or some streets in cities, lane lines are either not present or not sufficiently well signaled. In this work, we present a vision-based method to locate a vehicle within the road when no lane lines are present using only RGB images as input. To this end, we propose to fuse together the outputs of a semantic segmentation and a monocular depth estimation architecture to reconstruct locally a semantic 3D point cloud of the viewed scene. We only retain points belonging to the road and, additionally, to any kind of fences or walls that might be present right at the sides of the road. We then compute the width of the road at a certain point on the planned trajectory and, additionally, what we denote as the fence-to-fence distance. Our system is suited to any kind of motoring scenario and is especially useful when lane lines are not present on the road or do not signal the path correctly. The additional fence-to-fence distance computation is complementary to the road's width estimation. We quantitatively test our method on a set of images featuring streets of the city of Munich that contain a road-fence structure, so as to compare our two proposed variants, namely the road's width and the fence-to-fence distance computation. In addition, we also validate our system qualitatively on the Stuttgart sequence of the publicly available Cityscapes dataset, where no fences or walls are present at the sides of the road, thus demonstrating that our system can be deployed in a standard city-like environment. For the benefit of the community, we make our software open source. [ABSTRACT FROM AUTHOR]

Typically, lane departure warning systems rely on lane lines being present on the road.However, in many scenarios, e.g., secondary roads or some streets in cities, lane lines are eithernot present or not sufficiently well signaled. In this work, we present a vision-based method tolocate a vehicle within the road when no lane lines are present using only RGB images as input.To this end, we propose to fuse together the outputs of a semantic segmentation and a monoculardepth estimation architecture to reconstruct locally a semantic 3D point cloud of the viewed scene.We only retain points belonging to the road and, additionally, to any kind of fences or walls thatmight be present right at the sides of the road. We then compute the width of the road at a certainpoint on the planned trajectory and, additionally, what we denote as the fence-to-fence distance.Our system is suited to any kind of motoring scenario and is especially useful when lane lines arenot present on the road or do not signal the path correctly. The additional fence-to-fence distancecomputation is complementary to the road's width estimation. We quantitatively test our methodon a set of images featuring streets of the city of Munich that contain a road-fence structure, so asto compare our two proposed variants, namely the road's width and the fence-to-fence distancecomputation. In addition, we also validate our system qualitatively on the Stuttgart sequence of thepublicly available Cityscapes dataset, where no fences or walls are present at the sides of the road,thus demonstrating that our system can be deployed in a standard city-like environment. For thebenefit of the community, we make our software open source.