Search

While vision is an attractive alternative to many sensors targeting closed-loop controllers, it comes with high time-varying workload and robustness issues when targeted to edge devices with limited energy, memory and computing resources. Replacing classical vision processing pipelines, e.g., lane detection using Sobel filter, with deep learning algorithms is a way to deal with the robustness issues while hardware-efficient implementation is crucial for their adaptation for safe closed-loop systems. However, while implemented on an embedded edge device, the performance of these algorithms highly depends on their mapping on the target hardware and situation encountered by the system. That is, first, the timing performance numbers (e.g., latency, throughput) depends on the algorithm schedule, i.e., what part of the AI workload runs where (e.g., GPU, CPU) and their invocation frequency (e.g., how frequently we run a classifier). Second, the perception performance (e.g., detection accuracy) is heavily influenced by the situation - e.g., snowy and sunny weather condition provides very different lane detection accuracy. These factors directly influence the closed-loop performance, for example, the lane-following accuracy in a lane-keep assist system (LKAS). We propose a hardware- and situation-aware design of AI perception where the idea is to define the situations by a set of relevant environmental factors (e.g., weather, road etc. in an LKAS). We design the learning algorithms and parameters, overall hardware mapping and its schedule taking the situation into account. We show the effectiveness of our approach considering a realistic LKAS case-study on heterogeneous NVIDIA AGX Xavier platform in a hardware-in-the-loop framework. Our approach provides robust LKAS designs with 32% better performance compared to traditional approaches.

While vision is an attractive alternative to many sensors targeting closed-loop controllers, it comes with high time-varying workload and robustness issues when targeted to edge devices with limited energy, memory and computing resources. Replacing classical vision processing pipelines, e.g., lane detection using Sobel filter, with deep learning algorithms is a way to deal with the robustness issues while hardware-efficient implementation is crucial for their adaptation for safe closed-loop systems. However, while implemented on an embedded edge device, the performance of these algorithms highly depends on their mapping on the target hardware and situation encountered by the system. That is, first, the timing performance numbers (e.g., latency, throughput) depends on the algorithm schedule, i.e., what part of the AI workload runs where (e.g., GPU, CPU) and their invocation frequency (e.g., how frequently we run a classifier). Second, the perception performance (e.g., detection accuracy) is heavily influenced by the situation - e.g., snowy and sunny weather condition provides very different lane detection accuracy. These factors directly influence the closed-loop performance, for example, the lane-following accuracy in a lane-keep assist system (LKAS). We propose a hardware- and situation-aware design of AI perception where the idea is to define the situations by a set of relevant environmental factors (e.g., weather, road etc. in an LKAS). We design the learning algorithms and parameters, overall hardware mapping and its schedule taking the situation into account. We show the effectiveness of our approach considering a realistic LKAS case-study on heterogeneous NVIDIA AGX Xavier platform in a hardware-in-the-loop framework. Our approach provides robust LKAS designs with 32% better performance compared to traditional approaches.

In recent years autonomous driving is becoming a hot topic with the continuous development of deep learning, computer vision and sensor technology. It comes with robustness and quality of control issues when an Image-Based Control system is targeted to edge devices with limited energy, memory and computing resources. Using end-to-end CNN-based approaches can improve robustness, however, will increase runtime and result in a low frame per second. On the other hand, the traditional hardware-efficient approaches are lack of situation-awareness. Different environmental factors (e.g. road layouts, types of lane markers and weather) have a great impact on lane detection accuracy, and thus influence the quality of control. As a result, traditional approaches can not ensure robustness in the real world. In this work, we propose a hardware- and situation-aware sensing method to a lane-keeping assist system with the traditional lane detection algorithm to make the system both hardware-efficient and robust. We define situations based on different features and identify them using light-weight CNN-based situation classifiers. Depending on the current situation, we dynamically configure the system knobs based on hardware- and situation-aware characterization. To show the effectiveness of our approach, we consider a hardware-in-the-loop framework on NVIDIA AGX Xavier platform. Besides, Webots is used as the simulation environment on the server. Our results show that the robustness is highly improved comparing to traditional approaches. Moreover, the quality of control is also 32% better due to the approximated image processing signal pipeline and predefined invocation scheme.

Using inelastic X-ray scattering beyond the dipole limit and hard X-ray photoelectron spectroscopy we establish the dual nature of the U 5f electrons in UM2Si2 (M = Pd, Ni, Ru, Fe), regardless of their degree of delocalization. We have observed that the compounds have in common a local atomic-like state that is well described by the U 5f(2) configuration with the Gamma((1))(1) and Gamma(2) quasi-doublet symmetry. The amount of the U 5f(3) configuration, however, varies considerably across the UM2Si2 series, indicating an increase of U 5f itineracy in going from M = Pd to Ni to Ru and to the Fe compound. The identified electronic states explain the formation of the very large ordered magnetic moments in UPd2Si2 and UNi2Si2, the availability of orbital degrees of freedom needed for the hidden order in URu2Si2 to occur, as well as the appearance of Pauli paramagnetism in UFe2Si2. A unified and systematic picture of the UM2Si2 compounds may now be drawn, thereby providing suggestions for additional experiments to induce hidden order and/or superconductivity in U compounds with the tetragonal body-centered ThCr2Si2 structure.

Using inelastic X-ray scattering beyond the dipole limit and hard X-ray photoelectron spectroscopy we establish the dual nature of the U 5f electrons in UM2Si2 (M = Pd, Ni, Ru, Fe), regardless of their degree of delocalization. We have observed that the compounds have in common a local atomic-like state that is well described by the U 5f(2) configuration with the Gamma((1))(1) and Gamma(2) quasi-doublet symmetry. The amount of the U 5f(3) configuration, however, varies considerably across the UM2Si2 series, indicating an increase of U 5f itineracy in going from M = Pd to Ni to Ru and to the Fe compound. The identified electronic states explain the formation of the very large ordered magnetic moments in UPd2Si2 and UNi2Si2, the availability of orbital degrees of freedom needed for the hidden order in URu2Si2 to occur, as well as the appearance of Pauli paramagnetism in UFe2Si2. A unified and systematic picture of the UM2Si2 compounds may now be drawn, thereby providing suggestions for additional experiments to induce hidden order and/or superconductivity in U compounds with the tetragonal body-centered ThCr2Si2 structure.

In recent years autonomous driving is becoming a hot topic with the continuous development of deep learning, computer vision and sensor technology. It comes with robustness and quality of control issues when an Image-Based Control system is targeted to edge devices with limited energy, memory and computing resources. Using end-to-end CNN-based approaches can improve robustness, however, will increase runtime and result in a low frame per second. On the other hand, the traditional hardware-efficient approaches are lack of situation-awareness. Different environmental factors (e.g. road layouts, types of lane markers and weather) have a great impact on lane detection accuracy, and thus influence the quality of control. As a result, traditional approaches can not ensure robustness in the real world. In this work, we propose a hardware- and situation-aware sensing method to a lane-keeping assist system with the traditional lane detection algorithm to make the system both hardware-efficient and robust. We define situations based on different features and identify them using light-weight CNN-based situation classifiers. Depending on the current situation, we dynamically configure the system knobs based on hardware- and situation-aware characterization. To show the effectiveness of our approach, we consider a hardware-in-the-loop framework on NVIDIA AGX Xavier platform. Besides, Webots is used as the simulation environment on the server. Our results show that the robustness is highly improved comparing to traditional approaches. Moreover, the quality of control is also 32% better due to the approximated image processing signal pipeline and predefined invocation scheme.

The second-order phase transition into a hidden order phase in URu2Si2 goes along with an order parameter that is still a mystery, despite 30 years of research. However, it is understood that the symmetry of the order parameter must be related to the symmetry of the low-lying local electronic f-states. Here, we present results of a spectroscopic technique, namely core-level nonresonant inelastic X-ray scattering (NIXS). This method allows for the measurement of local high-multipole excitations and is bulk-sensitive. The observed anisotropy of the scattering function unambiguously shows that the 5f ground-state wave function is composed mainly of the Gamma(1) with majority J(z) = |4 > + |4 > and/or Gamma(2) singlet states. The incomplete dichroism indicates the possibility that quantum states of other irreducible representation are mixed into the ground state.

The second-order phase transition into a hidden order phase in URu2Si2 goes along with an order parameter that is still a mystery, despite 30 years of research. However, it is understood that the symmetry of the order parameter must be related to the symmetry of the low-lying local electronic f-states. Here, we present results of a spectroscopic technique, namely core-level nonresonant inelastic X-ray scattering (NIXS). This method allows for the measurement of local high-multipole excitations and is bulk-sensitive. The observed anisotropy of the scattering function unambiguously shows that the 5f ground-state wave function is composed mainly of the Gamma(1) with majority J(z) = |4 > + |4 > and/or Gamma(2) singlet states. The incomplete dichroism indicates the possibility that quantum states of other irreducible representation are mixed into the ground state.