We revisit the idea of using deep neural networks for one-shot decoding random and structured codes, such as polar codes. Although it is possible to achieve maximum a posteriori (MAP) bit error rate (BER) performance both code families short codeword lengths, we observe that (i) codes are easier learn (ii) network able generalize codewords has never seen during training structured, but not These results provide some evidence can form algorithm, rather than only simple classifier. introduce...

The fusion of multimodal sensor streams, such as camera, lidar, and radar measurements, plays a critical role in object detection for autonomous vehicles, which base their decision making on these inputs. While existing methods exploit redundant information good environmental conditions, they fail adverse weather where the sensory streams can be asymmetrically distorted. These rare ``edge-case'' scenarios are not represented available datasets, architectures designed to handle them. To...

The training complexity of deep learning-based channel decoders scales exponentially with the codebook size and therefore number information bits. Thus, neural network decoding (NND) is currently only feasible for very short block lengths. In this work, we show that conventional iterative algorithm polar codes can be enhanced when sub-blocks decoder are replaced by (NN) based components. partition encoding graph into smaller train them individually, closely approaching maximum a posteriori...

Autonomous driving at level five does not only means self-driving in the sunshine. Adverse weather is especially critical because fog, rain, and snow degrade perception of environment. In this work, current state art light detection ranging (lidar) sensors are tested controlled conditions a fog chamber. We present problems disturbance patterns for four different lidar systems. Moreover, we investigate how tuning internal parameters can improve their performance bad situations. This great...

We present an imaging framework which converts three images from a gated camera into high-resolution depth maps with accuracy comparable to pulsed lidar measurements. Existing scanning systems achieve low spatial resolution at large ranges due mechanically-limited angular sampling rates, restricting scene understanding tasks close-range clusters dense sampling. Moreover, today's scanners suffer high cost, power consumption, form-factors, and they fail in the presence of strong backscatter....

Adverse weather conditions are very challenging for autonomous driving because most of the state-of-the-art sensors stop working reliably under these conditions. In order to develop robust and algorithms, tests with current in defined crucial determining impact bad each sensor. This work describes a testing evaluation methodology that helps benchmark novel sensor technologies compare them sensors. As an example, gated imaging is compared standard foggy It shown outperforms passive due...

Gated imaging is an emerging sensor technology for self-driving cars that provides high-contrast images even under adverse weather influence. It has been shown this can generate high-fidelity dense depth maps with accuracy comparable to scanning LiDAR systems. In work, we extend the recent Gated2Depth framework aleatoric uncertainty providing additional confidence measure estimates. This help filter out uncertain estimations in regions without any illumination. Moreover, show training on...

This work shows and analyzes the LiDAR performance in real-world heavy winter conditions captured Northern Europe. We review how low temperatures, salted roads turbulent snow front of a passenger car influence systems developed for automated driving functions. Two test cars were driven north Finland Sweden 1.5 weeks to gather large amount point cloud data different urban rural scenarios. show that benchmarked sensors have surprising differences winter. Some got mechanically frozen whereas...

This work introduces an evaluation benchmark for depth estimation and completion using high-resolution measurements with angular resolution of up to 25" (arcsecond), akin a 50 megapixel camera per-pixel available. Existing datasets, such as the KITTI benchmark, provide only sparse reference order magnitude lower - these are treated ground truth by existing methods. We propose methodology in four characteristic automotive scenarios recorded varying weather conditions (day, night, fog, rain)....

Environment perception for autonomous driving is doomed by the trade-off between range-accuracy and resolution: current sensors that deliver very precise depth information are usually restricted to low resolution because of technology or cost limitations. In this work, we exploit from an active gated imaging system based on cost-sensitive diode CMOS technology. Learning a mapping pixel intensities three slices produces super-resolved map image with respectable relative accuracy 5 % in 25-80...