Scene flow provides crucial motion information for autonomous driving. Recent LiDAR scene flow models utilize the rigid-motion assumption at the instance level, assuming objects are rigid bodies. However, these instance-level methods are not suitable for sparse radar point clouds. In this work, we present a novel Traffic-Aware Radar Scene-Flow (TARS) estimation method, which utilizes motion rigidity at the traffic level. To address the challenges in radar scene flow, we perform object detection and scene flow jointly and boost the latter. We incorporate the feature map from the object detector, trained with detection losses, to make radar scene flow aware of the environment and road users. From this, we construct a Traffic Vector Field (TVF) in the feature space to achieve holistic traffic-level scene understanding in our scene flow branch. When estimating the scene flow, we consider both point-level motion cues from point neighbors and traffic-level consistency of rigid motion within the space. TARS outperforms the state of the art on a proprietary dataset and the View-of-Delft dataset, improving the benchmarks by 23% and 15%, respectively.

Velocities measured by radar sensors suffer from ambiguities caused by aliasing effects associated with signal processing. These ambiguities are highly undesired as they affect otherwise very accurate speed measurements. For automotive applications utilizing radar sensors, this means that measured velocities are potentially unreliable which can lead to unsafe conditions for automated driving functionalities. This work presents the first approach to disambiguate radial velocities measured by radar sensors based on Deep Learning methods. By utilizing the presented method, radial velocity estimates can be obtained which are both highly accurate while being reliable as ambiguities are resolved.

Selecting an optimal neural network architecture tailored to a specific dataset is a time-consuming task due to numerous design possibilities. Neural Architecture Search (NAS) provides strategies to identify well performing networks in a limited timeframe. Zero-cost proxies offer a training-free approach to find potential architectures within a predefined search space. However, relying solely on these proxies often leads to unreliable results across diverse search spaces and datasets. In this paper, we present an empirical study of extended zero-cost proxies, termed few-cost proxies, obtained by training for a restricted number of epochs. Our analysis demonstrates that these few-cost proxies significantly enhance the ranking performance. Furthermore, novel few-cost proxies introduced in this study outperform previous methods significantly, achieving a Spearman correlation of 0.89 compared to the second-highest score of 0.847 on TSS-Cifar10, showcasing their effectiveness in the context of NAS.

Autonomous learning of robotic manipulation tasks is a promising approach to reduce manual engineering effort and increase flexibility in the future of industrial manufacturing. Although a lot of research has been done especially robotic assembly tasks requiring contact-rich compliant interaction remain a challenge for learning-based methods, since large amounts of interaction data are required. Incorporation of prior knowledge has long been seen as a possibility to make learning-based approaches tractable. The question is how can we enable process experts to encode their prior knowledge in grey-box models so that it can be used for learning robotic manipulation tasks? For that reason we propose a new grey-box learning approach, “Adaptive Manipulation Primitives” (AMP), introduced in this paper. AMPs combine compliant manipulation task specifications based on Manipulation Primitives Nets with Policy Gradient Reinforcement Learning. Our framework is evaluated in a real-world robotic assembly task. It is shown that learning to assemble industrial connector modules is possible with comparatively few real-world trials.

Autonomous learning of robotic manipulation tasks is a desirable proposition for the future of industrial manufacturing to increase flexibility and reduce manual engineering effort. In particular assembly tasks that require contact-rich manipulation skills are challenging to accomplish with classical robotic control methods. The Reinforcement Learning (RL) framework provides a possibility to learn complex behaviors based on interaction with the environment. Although a lot of research has been done robotic assembly tasks remain a challenge for pure learning-based systems. In this paper we give an overview on grey-box learning approaches that integrate prior knowledge and learning based methods. Different dimensions of knowledge injection are identified, and knowledge representations are described. These representations are discussed in the context of industrial assembly processes to answer the question: how can process experts model their knowledge to boost RL approaches in the context of industrial assembly?