Learning Deconvolution Network for Semantic Segmentation

We report recent results from field experiments conducted with a team of ground and aerial robots engaged in the collaborative mapping of an earthquake-damaged building. The goal of the experimental exercise is the generation of three-dimensional maps that capture the layout of a multifloor environment. The experiments took place in the top three floors of a structurally compromised building at Tohoku University in Sendai, Japan that was damaged during the 2011 Tohoku earthquake. We provide details of the approach to the collaborative mapping and report results from the experiments in the form of maps generated by the individual robots and as a team. We conclude by discussing observations from the experiments and future research topics. © 2012 Wiley Periodicals, Inc.

(This work builds upon the conference paper (Michael et al., 2012).)

/pdf/collaborative-mapping-of-an-earthquake-damaged-building-via-4ees2efish.pdf

Collaborative mapping of an earthquake-damaged building via ground and aerial robots

We present DeepVCP - a novel end-to-end learning-based 3D point cloud registration framework that achieves comparable registration accuracy to prior state-of-the-art geometric methods. Different from other keypoint based methods where a RANSAC procedure is usually needed, we implement the use of various deep neural network structures to establish an end-to-end trainable network. Our keypoint detector is trained through this end-to-end structure and enables the system to avoid the interference of dynamic objects, leverages the help of sufficiently salient features on stationary objects, and as a result, achieves high robustness. Rather than searching the corresponding points among existing points, the key contribution is that we innovatively generate them based on learned matching probabilities among a group of candidates, which can boost the registration accuracy. We comprehensively validate the effectiveness of our approach using both the KITTI dataset and the Apollo-SouthBay dataset. Results demonstrate that our method achieves comparable registration accuracy and runtime efficiency to the state-of-the-art geometry-based methods, but with higher robustness to inaccurate initial poses. Detailed ablation and visualization analysis are included to further illustrate the behavior and insights of our network. The low registration error and high robustness of our method make it attractive to the substantial applications relying on the point cloud registration task.

/pdf/deepvcp-an-end-to-end-deep-neural-network-for-point-cloud-w2mhnpx0xf.pdf

DeepVCP: An End-to-End Deep Neural Network for Point Cloud Registration

Autonomous vehicles (AV) are expected to improve, reshape, and revolutionize the future of ground transportation. It is anticipated that ordinary vehicles will one day be replaced with smart vehicles that are able to make decisions and perform driving tasks on their own. In order to achieve this objective, self-driving vehicles are equipped with sensors that are used to sense and perceive both their surroundings and the faraway environment, using further advances in communication technologies, such as 5G. In the meantime, local perception, as with human beings, will continue to be an effective means for controlling the vehicle at short range. In the other hand, extended perception allows for anticipation of distant events and produces smarter behavior to guide the vehicle to its destination while respecting a set of criteria (safety, energy management, traffic optimization, comfort). In spite of the remarkable advancements of sensor technologies in terms of their effectiveness and applicability for AV systems in recent years, sensors can still fail because of noise, ambient conditions, or manufacturing defects, among other factors; hence, it is not advisable to rely on a single sensor for any of the autonomous driving tasks. The practical solution is to incorporate multiple competitive and complementary sensors that work synergistically to overcome their individual shortcomings. This article provides a comprehensive review of the state-of-the-art methods utilized to improve the performance of AV systems in short-range or local vehicle environments. Specifically, it focuses on recent studies that use deep learning sensor fusion algorithms for perception, localization, and mapping. The article concludes by highlighting some of the current trends and possible future research directions.

https://hal.archives-ouvertes.fr/hal-02942600/document

Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review.

We propose a real-time, low-drift laser odometry approach that tightly integrates sequentially measured 3D multi-beam LIDAR data with inertial measurements. The laser measurements are motion-compensated using a novel algorithm based on non-rigid registration of two consecutive laser sweeps and a local map. IMU data is being tightly integrated by means of factor-graph optimization on a pose graph. We evaluate our method on a public dataset and also obtain results on our own datasets that contain information not commonly found in existing datasets. At the time of writing, our method was ranked within the top five laser-only algorithms of the KITTI odometry benchmark.

MC2SLAM: Real-Time Inertial Lidar Odometry Using Two-Scan Motion Compensation

Probabilistic terrain classification in unstructured environments

Three-dimensional laser range finders provide autonomous systems with vast amounts of information. However, autonomous robots navigating in unstructured environments are usually not interested in every geometric detail of their surroundings. Instead, they require real-time information about the location of obstacles and the condition of drivable areas.In this paper, we first present grid-based algorithms for classifying regions as either drivable or not. In a subsequent step, drivable regions are further examined using a novel algorithm which determines the local terrain roughness. This information can be used by a path planning algorithm to decide whether to prefer a rough, muddy area, or a plain street, which would not be possible using binary drivability information only.

/pdf/terrain-drivability-analysis-in-3d-laser-range-data-for-3r0mh0158r.pdf

Terrain drivability analysis in 3D laser range data for autonomous robot navigation in unstructured environments

Mobile systems for mapping and terrain classification are often tested on datasets of intact environments only. The behavior of the algorithms in unstructured environments is mostly unknown. In safety, security and rescue environments, the robots have to handle much rougher terrain. Therefore, there is a need for 3D test data that also contains disaster scenarios. During the Response Robot Evaluation Exercise in March 2010 in Disaster City, College Station, Texas (USA), a comprehensive dataset was recorded containing the data of a 3D laser range finder, a GPS receiver, an IMU and a color camera. We tested our algorithms (for terrain classification and 3D mapping) with the dataset, and will make the data available to give other researchers the chance to do the same. We believe that this captured data of this well known location provides a valuable dataset for the USAR robotics community, increasing chances of getting more comparable results.

/pdf/real-time-3d-mapping-of-rough-terrain-a-field-report-from-58mbfxyc1n.pdf

Real-time 3D mapping of rough terrain: A field report from Disaster City

When crafting driver assistance systems designed for truck/trailer combinations, knowledge about the position and orientation of a truck relative to the attached trailer is a vital prerequisite for any kinematic calculation and trajectory estimation. An advanced optical sensor system measuring the 3-D state of an attached two-axle trailer is proposed in this publication. It uses a Kalman filter for enhanced pose estimation and is evaluated against previous versions of the sensor system for the same purpose.

Frank Neuhaus

Papers

MC2SLAM: Real-Time Inertial Lidar Odometry Using Two-Scan Motion Compensation

Probabilistic terrain classification in unstructured environments

Terrain drivability analysis in 3D laser range data for autonomous robot navigation in unstructured environments

Real-time 3D mapping of rough terrain: A field report from Disaster City

Advanced 3-D trailer pose estimation for articulated vehicles