No abstract

The ability to detect objects regardless of image distortions or weather conditions is crucial for real-world applications of deep learning like autonomous driving. We here provide an easy-to-use benchmark to assess how object detection models perform when image quality degrades. The three resulting benchmark datasets, termed Pascal-C, Coco-C and Cityscapes-C, contain a large variety of image corruptions. We show that a range of standard object detection models suffer a severe performance loss on corrupted images (down to 30-60% of the original performance). However, a simple data augmentation trick-stylizing the training images-leads to a substantial increase in robustness across corruption type, severity and dataset. We envision our comprehensive benchmark to track future progress towards building robust object detection models. Benchmark, code and data are available at: http://github.com/bethgelab/robust-detection-benchmark

This paper presents an approach for cycle-accurate simulation of embedded software by integration in an abstract SystemC model. Compared to existing simulation-based approaches, we present a hybrid method that resolves performance issues by combining the advantages of simulation-based and analytical approaches. In a first step, cycle-accurate static execution time analysis is applied at each basic block of a cross-compiled binary program using static processor models. After that, the determined timing information is back-annotated into SystemC for fast simulation of all effects that can not be resolved statically. This allows the consideration of data dependencies during run-time and the incorporation of branch prediction and cache models by efficient source code instrumentation. The major benefit of our approach is that the generated code can be executed very efficiently on the simulation host with approximately 90% of the speed of the untimed software without any code instrumentation.

The virtual validation of automated driving functions requires meaningful simulation models of environment perception sensors such as radar, lidar, and cameras. There does not yet exist an unrivaled standard for perception sensor models, and radar especially lacks modeling approaches that consistently produce realistic results. In this paper, we present measurements that exemplify challenges in the development of meaningful radar sensor models. We highlight three major challenges: multi-path propagation, separability, and sensitivity of radar cross section to the aspect angle. We also review previous work addressing these challenges and suggest further research directions towards meaningful automotive radar simulation models.

We investigate a generic self-adaptation method to reduce the design effort for System-on-Chip (SoC). Previous self-adaptation solutions at chip-level use circuitries which have been specially designed for the current problem by hand, leading to an elaborate and inflexible design process, requiring specially trained engineers, and making design reuse difficult. On the other hand, a generic self-adaptation method that can be used for various self-adaptation problems promises to reduce the necessary design effort, but may come with reduced performance and other costs.In this paper, we analyze the performance, self-adaptation capabilities and costs of a generic self-adaptation method. The proposed method allows chip-level self-adaptation of a SoC, can tolerate unforeseen events, and can generalize from previous self-adaptation tasks. Furthermore, the method helps to improve the design process by allowing design reuse, providing generic applicability, and offering a uniform design process for various self-adaptation tasks. Simulation results show that the performance of our method lies only 10% below the performance of a perfect, non-adaptive system in the average case, and only 32% in the worst case. In case of unforeseen events, where the performance of a non-adaptive system decreases significantly, the method can keep its performance level by self-adaptation. We also compare other costs involved.

This paper presents an approach for accurately estimating the execution time of parallel software components in complex embedded systems. Timing annotations obtained from highly optimized binary code are added to the source code of software components which is then integrated into a SystemC transaction-level simulation. This approach allows a fast evaluation of software execution times while being as accurate as conventional instruction set simulators. By simulating binary-level control flow in parallel to the original functionality of the software, even compiler optimizations heavily modifying the structure of the generated code can be modeled accurately. Experimental results show that the presented method produces timing estimates within the same level of accuracy as an established commercial tool for cycleaccurate instruction set simulation while being at least 20 times faster.

No abstract

International audienceThis article presented a survey of dependability issues faced by multi-core architectures at nanoscale technology nodes. Existing solutions against these challenges were also discussed, describing their scope of application, from technology level methodologies, to design approaches to the metrics required to evaluate the overall dependability of a system. In the future, the constant reduction of the feature size of the devices will exacerbate the issues related to aging and soft errors. This will create further challenges and at design level, an integrated design approach that will cope with the occurrence of faults at any time of their occurrence i.e., at manufacturing (thus increasing yield) and in the field (thus increasing reliability) will become more and more important to obtain economically viable and dependable systems. Dependability assessment will also need an integrated approach for cross-layer, pre- and post-silicon techniques for “just right”dependability assessment in order to avoid “overdesign”for dependability using classic guard-banding methodologies