Thanks to the efforts of the robotics and autonomous systems community, the myriad applications and capacities of robots are ever increasing. There is increasing demand from end users for autonomous service robots that can operate in real environments for extended periods. In the Spatiotemporal Representations and Activities for Cognitive Control in Long-Term Scenarios (STRANDS) project (http://strandsproject.eu), we are tackling this demand head-on by integrating state-of-the-art artificial intelligence and robotics research into mobile service robots and deploying these systems for long-term installations in security and care environments. Our robots have been operational for a combined duration of 104 days over four deployments, autonomously performing end-user-defined tasks and traversing 116 km in the process. In this article, we describe the approach we used to enable long-term autonomous operation in everyday environments and how our robots are able to use their long run times to improve their own performance.

In this letter, we present a deep learning-based network, GCNv2, for generation of keypoints and descriptors. GCNv2 is built on our previous method, GCN, a network trained for 3D projective geometry. GCNv2 is designed with a binary descriptor vector as the ORB feature so that it can easily replace ORB in systems such as ORB-SLAM2. GCNv2 significantly improves the computational efficiency over GCN that was only able to run on desktop hardware. We show how a modified version of ORBSLAM2 using GCNv2 features runs on a Jetson TX2, an embedded low-power platform. Experimental results show that GCNv2 retains comparable accuracy as GCN and that it is robust enough to use for control of a flying drone. Source code is available at: https://github.com/jiexiong2016/GCNv2_SLAM.

This paper describes a system for reacquiring features of interest in a shallow-water ocean environment, using autonomous underwater vehicles (AUVs) equipped with low-cost sonar and navigation sensors. In performing mine countermeasures, it is critical to enable AUVs to navigate accurately to previously mapped objects of interest in the water column or on the seabed, for further assessment or remediation. An important aspect of the overall system design is to keep the size and cost of the reacquisition vehicle as low as possible, as it may potentially be destroyed in the reacquisition mission. This low-cost requirement prevents the use of sophisticated AUV navigation sensors, such as a Doppler velocity log (DVL) or an inertial navigation system (INS). Our system instead uses the Proviewer 900-kHz imaging sonar from Blueview Technologies, which produces forward-looking sonar (FLS) images at ranges up to 40 mat approximately 4 Hz. In large volumes, it is hoped that this sensor can be manufactured at low cost. Our approach uses a novel simultaneous localization and mapping (SLAM) algorithm that detects and tracks features in the FLS images to renavigate to a previously mapped target. This feature-based navigation (FBN) system incorporates a number of recent advances in pose graph optimization algorithms for SLAM. The system has undergone extensive field testing over a period of more than four years, demonstrating the potential for the use of this new approach for feature reacquisition. In this report, we review the methodologies and components of the FBN system, describe the system's technological features, review the performance of the system in a series of extensive in-water field tests, and highlight issues for future research.

We describe an approach to simultaneous localization and mapping, SLAM. This approach has the highly desirable property of robustness to data association errors. Another important advantage of our algorithm is that non-linearities are computed exactly, so that global constraints can be imposed even if they result in large shifts to the map. We represent the map as a graph and use the graph to find an efficient map update algorithm. We also show how topological consistency can be imposed on the map, such as, closing a loop. The algorithm has been implemented on an outdoor robot and we have experimental validation of our ideas. We also explain how the graph can be simplified leading to linear approximations of sections of the map. This reduction gives us a natural way to connect local map patches into a much larger global map.

Autonomous vehicles rely on GPS aided by motion sensors to localize globally within the road network. However, not all driving surfaces have satellite visibility. Therefore, it is important to augment these systems with localization based on environmental sensing such as cameras, lidar and radar in order to increase reliability and robustness. In this work we look at using radar for localization. Radar sensors are available in compact format devices well suited to automotive applications. Past work on localization using radar in automotive applications has been based on careful sensor modeling and Sequential Monte Carlo, (Particle) filtering. In this work we investigate the use of the Iterative Closest Point, ICP, algorithm together with an Extended Kalman filter, EKF, for localizing a vehicle equipped with automotive grade radars. Experiments using data acquired on public roads shows that this computationally simpler approach yields sufficiently accurate results on par with more complex methods.

In this paper, we propose a new learning scheme for generating geometric correspondences to be used for visual odometry. A convolutional neural network (CNN) combined with a recurrent neural network (RNN) are trained together to detect the location of keypoints as well as to generate corresponding descriptors in one unified structure. The network is optimized by warping points from source frame to reference frame, with a rigid body transform. Essentially, learning from warping. The overall training is focused on movements of the camera rather than movements within the image, which leads to better consistency in the matching and ultimately better motion estimation. Experimental results show that the proposed method achieves better results than both related deep learning and hand crafted methods. Furthermore, as a demonstration of the promise of our method we use a naive SLAM implementation based on these keypoints and get a performance on par with ORB-SLAM.