Place recognition can be achieved by identifying whether a pair of images (a labeled reference image and a query image) depict the same place, regardless of appearance changes due to different viewpoints or lighting conditions. It is an important component of systems for camera localization and for loop closure detection and a widely studied problem for indoor or urban environments. Recently, the use of robots in agriculture and automatic gardening has created new challenges due to the highly repetitive appearance with prevalent green color and repetitive texture of garden-like scenes. The lack of available data recorded in gardens or plant fields makes difficult to improve localization algorithms for such environments. In this paper, we propose a new data set of garden images for testing algorithms for visual place recognition. It contains images with ground truth camera pose recorded in real gardens at different times, with varying light conditions. We also provide ground truth for all possible pairs of images, indicating whether they depict the same place or not. We also performed a thorough benchmark of several holistic (whole-image) descriptors, and provide the results on the proposed data set. We observed that existing descriptors have difficulties with scenes with repetitive textures and large changes of camera viewpoint.

Robots are increasingly present in modern industry and also in everyday life. Their applications range from health-related situations, for assistance to elderly people or in surgical operations, to automatic and driver-less vehicles (on wheels or flying) or for driving assistance. Recently, an interest towards robotics applied in agriculture and gardening has arisen, with applications to automatic seeding and cropping or to plant disease control, etc. Autonomous lawn mowers are succesful market applications of gardening robotics. In this paper, we present a novel robot that is developed within the TrimBot2020 project, funded by the EU H2020 program. The project aims at prototyping the first outdoor robot for automatic bush trimming and rose pruning.

In this work an approach to generate collisions-free trajectories for vessels is presented. The approach consist of a global path planner and a local evasive trajectory planner. The global path planner uses chart information from ENCs to estimate waypoints for a global path avoiding obstacles defined in a chart. The local evasive trajectory planner uses information of a target tracking system to proof for a subsection of the global path if a collisions with other vessels will probably occur. In case of possible collisions the local evasive trajectory planner estimates an evasive trajectory to avoid the dynamic obstacles and lead back to the global path after the avoidance manoeuvre. For the global path planning algorithm a grid based reachability graph is applied to consider the turning circle of the own vessel. The local planner uses a grid based search algorithm that takes into account the turning circle of the vessel also. The result of both search algorithms is a sequence of waypoints defining a collision free path. A Bézier curve interpolation is applied to these waypoints to achieve a path with continuous curvatures suitable for path following algorithms.

In this paper, an approach for a collision avoidance system for small Unmanned Surface Vehicles (USV) is presented. These kind of USVs are typically used for measuring tasks and thus, the collision avoidance algorithm has to take the mission objective and information about obstacles and other vessels into account. It generates an evasive trajectory that either leads around the obstacle, reduces the velocity to let another vessel pass or combines evasive manoeuvres and velocity changes. For the evasive path generation, a specialized A* algorithm is used that allows velocity variations and considers different turning circles for different velocities. It is applied to a grid representation of the environment that is built using a laser range finder. The laser range finder detects obstacles above the surface like other vessels, pillars or jetties. An additional tracking module is used to determine the dynamic state of moving obstacles such as other vessels. The complete collision avoidance system is integrated into a small USV and results of real world tests for environment mapping and target tracking are presented. The collision avoidance performance is shown for simulated measurement scenarios.

The improvement of collision avoidance for vessels in close range encounter situations is an important topic for maritime traffic safety. Typical approaches generate evasive trajectories or optimise the trajectories of all involved vessels. Such a collision avoidance system has to produce evasive manoeuvres that do not confuse other navigators. To achieve this behaviour, a probabilistic obstacle handling based on information from a radar sensor with target tracking, that considers measurement and tracking uncertainties is proposed. A grid based path search algorithm, that takes the information from the probabilistic obstacle handling into account, is then used to generate evasive trajectories. The proposed algorithms have been tested and verified in a simulated environment for inland waters.

The improvement of collision avoidance for vessels in close range encounter situations is an important topic for maritime traffic safety. Typical approaches generate evasive trajectories or optimise the trajectories of all involved vessels. The idea of this work is to validate these trajectories related to guaranteed motion safety, which means that it is not sufficient for a trajectory to be collision-free, but it must additionally ensure that an evasive manoeuvre is performable at any time. An approach using the distance and the evolution of the distance to the other vessels is proposed. The concept of Inevitable Collision States (ICS) is adopted to identify the states for which no evasive manoeuvre exist. Furthermore, it is implemented into a collision avoidance system for recreational crafts to demonstrate the performance.

The problem of vessel collisions or near-collision situations on sea, often caused by human error due to incomplete or overwhelming information, is becoming more and more important with rising maritime traffic. Approaches to supply navigators and Vessel Traffic Services with expert knowledge and suggest trajectories for all vessels to avoid collisions, are often aimed at situations where a single planner guides all vessels with perfect information. In contrast, the authors suggest a two-part procedure which plans trajectories using a specialised A* and negotiates trajectories until a solution is found, which is acceptable for all vessels. The solution obeys collision avoidance rules, includes a dynamic model of all vessels and negotiates trajectories to optimise globally without a global planner and extensive information disclosure. The procedure combines all components necessary to solve a multi-vessel encounter and is tested currently in simulation and on several test beds. The first results show a fast converging optimisation process which after a few negotiation rounds already produce feasible, collision free trajectories.

Collision avoidance for vessels highly depends on a robust obstacle detection. This is commonly achieved by use of high precision radar sensing. For smaller vessels however, the use of low-cost sensors is typical. The idea of this work is to improve the robustness of collision avoidance by integrating a sensor model together with the collision avoidance algorithm in order to consider the accuracy of the measurements. Furthermore, a target tracking algorithm based on an interacting Multi-Model Filter (IMM) is used for robust obstacle detection.

This paper presents a method to efficiently solve the control allocation problem of vessels propelled by two azimuth thrusters with limited panning range. The use of rotatable thrusters results in a configuration matrix that, in the context of control allocation, typically leads to a constrained nonlinear optimization problem. It is shown, how this can be transformed into an easy to solve scalar unconstrained problem by using simple geometrical relations. Using a cost criterion from the literature for minimum power consumption, however, leads to discontinuities due to the limited panning angle range of the thrusters. By introducing a parameterized cost function, the discontinuities in the panning motion of the thrusters can be largely avoided.

In this paper, the process and results of a parameter identification task for a maritime unmanned surface vehicle are presented. The system has been modeled using state of the art methodology, and the parameters have been estimated via a weighted least square optimization approach. The required measurement data have been taken from various maneuvre trials. For solving the optimization problem, a Particle Swarm programming approach has been used, which has reliably found the global minimum of the cost function. The results show excellent agreement between measured and simulated data.

This paper provides an overview of the development of a catamaran-like autonomous surface vessel. The vessel is intended to serve as an experimental platform to study algorithms for control, sensor data fusion as well as path planning and collision avoidance. The mechanical design, the electronics and software architecture is described together with the sensor and actuator setup. The propulsion system comprises of two azimuth thrusters with limited azimuth angles. This concept provides additional degrees of freedom resulting in an overactuated control system. A mathematical model of the vessel has been adapted and implemented for model based control strategies, simulation and hardware in the loop testing. In this paper some results of the parameter identification process are presented. For the overactuated system a control allocation algorithm has been developed.

In this work a grid based collision avoidance algorithm which considers the physical constrains of a vessel is presented. For this purpose a new geometry neighbourhood is introduced and explained in detail. The collision avoidance algorithm pays attention to the COLREGs and provides a collision-free path. To find this path, Lee's algorithm is used.

In this work a real-time grid based collision avoidance algorithm for vessels in maritime environments is presented. Most grid based collision avoidance approaches use Lee's algorithm to find an optimal collision-free path. The drawback of this algorithm is the long runtime for large grids. To reduce this runtime, in this work the A* search is used instead of Lee's algorithm. The algorithm additionally regards the physical constrains of the vessel and pays attention to the COLREGs. The runtime and the resulting path of Lee's algorithm and the A* search are compared in computational experiments. Furthermore, the performance of the collision avoidance on a real vessel on the Lake Constance is presented.

This paper presents a mapping approach for inland waters using a noisy radar sensor installed on a boat. The vessel’s position is acquired from GPS, thus this is a pure mapping problem. For the actual mapping the probabilistic open-source mapping framework octomap as presented by [8] is used. Exactly one polygon is extracted from a binary radar image, the so-called Water Enclosing Polygon. This discards inland echos and multi-path measurements. Additionally, an approach to detect bridges and dolphins is presented. The runtime of the mapping algorithm is less then 2.5 s. Thus, each new radar scan is integrated into the octomap.

Detecting objects is of fundamental importance for the Eurobot Challenge 2011. This paper presents a Kinect-based approach to detect the game elements on the game field. Using the Kinect sensor provides the advantage that elements lying behind other elements can still be detected, which is nearly impossible for a laser-based approach. The Kinect provides depth information which is projected to the 3D space, building a point cloud of the game elements. The point cloud is then analyzed for the clusters of the game elements which are passed to a classifier each. The classifier decides if the passed-in cluster is a pawn, king or the enemy robot.

For many indoor robot applications it is crucial to recognize doors reliably. Doors connect rooms and hallways and therefore determine possible navigation paths. Moreover, doors are important landmarks for robot self-localization and map building. Existing algorithms for door detection are often limited to restricted environments. They do not consider the large intra-class variability of doors. In this paper we present an approach which combines a set of simple door detection classifiers. The classifiers are based either on visual information or on laser ranger data. Separately, these classifiers accomplish only a weak door detection rate. However, by combining them with the aid of AdaBoost Algorithm more than 82% of all doors with a false positive rate less than 3% are detected in static test data. Further improvement can easily be achieved by using different door perspectives from a moving robot. In a realtime mobile robot application we detect more than 90% of all doors with a very low false detection rate.

During the last few years mobile robots got more and more important to solve different tasks in outdoor and indoor environments. To solve these tasks one very essential issue is to get from one point A to another point B as fast as possible. To find the least expensive route to the goal the pathfinding process needs a full state space information about the environment. With this information we can use optimally efficient algorithms like A* to find the route, but this might be very expensive on memory usage and time response. therefore we need to use other data structures to represent the whole information about the environment. This paper shows how the usage of quadtrees improves performance in terms of computation speed, memory requirements and path length.

In this paper the design of an omni-directional sensor which measures the angles between active infrared beacons is proposed. The aim is to develop a simple and inexpensive sensor and beacon system. Therefore an array of ten CMOS cameras with an infrared filter is used. We present the design steps and the evaluation of the sensor in detail. The systematic errors of the system are analyzed during the evaluation of the sensor's accuracy. They are compensated by a linear regression, which leads to an accuracy of less than one degree. The achieved accuracy qualifies the system for applications like landmark-based localization.

The objective of this work is to propose a landmark based localization system for the Eurobot contest, which enhance the positioning accuracy compared to an odometry based localization. To detect the landmarks a visual sensor which measures the angle between the robot and these landmarks is used. Based on these measurements two approaches to determinate the robot’s position are presented: a triangulation method and an extended Kalman filter (EKF) approach. The extended Kalman filter approach combines the landmark measurements and the odometry. A robot from the Eurobot 2010 is used to carry out experimental results for the EKF based localization and present the enhancement of this approach.

Doors are important landmarks for robot self localization and navigation in indoor environments. Existing algorithms for door detection are often limited to restricted environments. They do not consider the large intra-class variability of doors. In this paper we present a camera- and laser-based approach which allows finding more than 82% of all doors with a false positive rate less than 3% in static test sets. By using different door perspectives from a moving robot, we detect more than 90% of doors with a very low false detection rate.