Visibility is a fundamental concept in computational geometry, with numerous applications in surveillance, robotics, and games. This software paper presents TřiVis, a C++ library developed by the authors for computing numerous visibility-related queries in highly complex polygonal environments. Adapting the triangular expansion algorithm, TřiVis stands out as a versatile, high-performance, more reliable and easy-to-use alternative to current solutions that is also free of heavy dependencies. Through evaluation on a challenging dataset, TřiVis has been benchmarked against existing visibility libraries. The results demonstrate that TřiVis outperforms the competing solutions by at least an order of magnitude in query times, while exhibiting more reliable runtime behavior. TřiVis is freely available for private, research, and institutional use at https://github.com/janmikulacz/trivis.

Visual classification under uncertainty is a complex computer vision problem. We present a thorough comparison of several variants of convolutional neural network (CNN) classification techniques in the context of ambiguous image data interpretation. We explore possible improvements in classification accuracy achieved by insertion of prior ambiguity information during the annotation process. This enables us to harness known similarities between individual classes and use them as probability distributions for soft ground-truth labels. We also present an approach based on Bayesian CNNs, offering the possibility of further interpretation of classification results in a problem where the neural network model is often considered as a black box. The presented techniques are verified on a practical spot weld inspection problem.

We revise, improve and extend the system previously introduced by us and named SSM-VPR (Semantic and Spatial Matching Visual Place Recognition), largely boosting its performance above the current state of the art. The system encodes images of places by employing the activations of different layers of a pre-trained, off-the-shelf, VGG16 Convolutional Neural Network (CNN) architecture. It consists of two stages: given a query image of a place, (1) a list of candidates is selected from a database of places and (2) the candidates are geometrically compared with the query. The comparison is made by matching CNN features and, equally important, their spatial locations, selecting the best candidate as the recognized place. The performance of the system is maximized by finding optimal image resolutions during the second stage and by exploiting temporal correlation between consecutive frames in the employed datasets.

We propose a novel teach-and-repeat navigation system, SSM-Nav, which is based on the output of the recently introduced SSM visual place recognition methodology. During the teach phase, a teleoperated wheeled robot stores in a database features of images taken along an arbitrary route. During the repeat phase or navigation, a CNN-based comparison of each captured image is performed against the database. With the help of a particle filter, the image associated with the most likely location is selected at each time and its horizontal offset with respect to the current scene used to correct the steering of the robot and to navigate. Indoor tests in our lab show a maximum error of less than 10cm and excellent robustness to perturbations such as drastic changes in illumination, lateral displacements, different starting positions, or even kidnapping. Preliminary outdoor tests on a 0.22km route show promising results, with an estimated maximum error of less than 25cm.

In a robotised warehouse, as in any place where robots move autonomously, a major issue is the localization or detection of human operators during their intervention in the work area of the robots. This paper introduces a wearable human localization system for large warehouses, which utilize preinstalled infrastructure used for localization of automated guided vehicles (AGVs). A monocular down-looking camera is detecting ground nodes, identifying them and computing the absolute position of the human to allow safe cooperation and coexistence of humans and AGVs in the same workspace. A virtual safety area around the human operator is set up and any AGV in this area is immediately stopped. In order to avoid triggering an emergency stop because of the short distance between robots and human operators, the trajectories of the robots have to be modified so that they do not interfere with the human. The purpose of this paper is to demonstrate an absolute visual localization method working in the challenging environment of an automated warehouse with low intensity of light, massively changing environment and using solely monocular camera placed on the human body.

This paper addresses the problem of coordination of a fleet of mobile robots - the problem of finding an optimal set of collision-free trajectories for individual robots in the fleet. Many approaches have been introduced during the last decades, but a minority of them is practically applicable, i.e. fast, producing near-optimal solutions, and complete. We propose a novel probabilistic approach based on the Rapidly Exploring Random Tree algorithm (RRT) by significantly improving its multi-robot variant for discrete environments. The presented experimental results show that the proposed approach is fast enough to solve problems with tens of robots in seconds. Although the solutions generated by the approach are slightly worse than one of the best state-of-the-art algorithms presented in (ter Mors et al., 2010), it solves problems where ter Mors's algorithm fails.

Motion planning techniques suited for initialization of Model Predictive Control based methodology applied for complex maneuvering and stabilization of formations of Micro Aerial Vehicles are proposed in this paper. Two approaches to initialization of the formation driving method will be described, experimentally verified, evaluated and compared. The first proposed method is based on multiobjective optimization of the trajectory guess obtained by a Rapidly Exploring Random Trees technique. It represents an easy to implement and robust method suited for off-line initialization of the formation driving algorithm. The second proposed method is based on sequential processing of parts of the obtained trajectory. This method is well scalable and thus applicable in large workspaces with complex obstacles. In addition, the second method enables a significant reduction of computational time as is shown by comparison of series of simulations in different environments.

In this article we introduce the problem of finding an optimal path in order to find a stationary object placed in the environment whose map is not a-priory known. At first sight the problem seems to be similar to exploration which has been thoroughly studied by the robotic community. We show that a general framework for search can be derived from frontier-based exploration, but exploration strategies for selection of a next goal to which navigate a robot cannot be simply reused. We present three goal selection strategies (greedy, traveling salesmen based, and traveling deliveryman based) and statistically evaluate and discuss their performance for search in comparison to exploration.

A stabilization and control technique developed for steering swarms of unmanned micro aerial vehicles is proposed in this paper. The presented approach based on a visual relative localization of swarm particles is designed for utilization of multi-robot teams in real-world dynamic environments. The core of the swarming behaviour is inspired by Reynold's BOID model proposed for 2D simulations of schooling behaviour of fish. The idea of the simple BOID model, with three simple rules: Separation, Alignment and Cohesion, is extended for swarms of quadrotors in this paper. The proposed solution integrates the swarming behaviour with the relative localization and with a stabilization and control mechanism, which respects fast dynamics of unmanned quadrotors. The proposed method aspires to be an enabling technique for deployment of swarms of micro areal vehicles outside laboratories that are equipped with precise positioning systems. The swarming behaviour as well as the possibility of swarm stabilization with the visual relative localization in the control feedback are verified by simulations and partly by an experiment with quadrotors in this paper.

Modular robots, which are systems made of many robotic modules, can utilize various types of locomotion. Different approaches can be used to generate these basic motion skills — motion primitives. To move in a complex environment, several motion primitives are needed and a mechanism to switch them is required. This can be realized using a high-level motion planning. To enable autonomous operation of modular robots equipped with limited computational resources, it is necessary to generate the motion plans on-board, i.e., without external computers. In this paper, we propose a novel simplified motion model of a modular robot, which allows the robot to employ the motion planner as a fast on-board replanner. The proposed approach has been verified both in simulations as well as with real robots.

In this paper, we present a small, light-weight, low-cost, fast and reliable system designed to satisfy requirements of relative localization within a swarm of micro aerial vehicles. The core of the proposed solution is based on off-the-shelf components consisting of the Caspa camera module and Gumstix Overo board accompanied by a developed efficient image processing method for detecting black and white circular patterns. Although the idea of the roundel recognition is simple, the developed system exhibits reliable and fast estimation of the relative position of the pattern up to 30 fps using the full resolution of the Caspa camera. Thus, the system is suited to meet requirements for a vision based stabilization of the robotic swarm. The intent of this paper is to present the developed system as an enabling technology for various robotic tasks.

The ability to move in complex environments is a key property required for deployment of modular robots in challenging applications like search & rescue missions or space exploration. Wide range of motion types like crawling or walking can be achieved using Central Pattern Generators producing periodic control signals. Although these motions can be very effective to steer robots in their vicinity or in a given direction, they need to be switched to reach a far position in the environment. This paper presents a novel modification of Rapidly Exploring Random Tree (RRT) algorithm for modular robots. For efficient exploration of the configuration space, predefined motion primitives are used. While the motion primitives provide effective local motions, the RRT-based planner switches them in order to reach the desired global goal.

This paper presents a multi-goal path planning framework based on a self-organizing map algorithm and a model of the navigation describing evolution of the localization error. The framework combines finding a sequence of goals' visits with a goal-to-goal path planning considering localization uncertainty. The approach is able to deal with local properties of the environment such as expected visible landmarks usable for the navigation. The local properties affect the performance of the navigation, and therefore, the framework can take the full advantage of the local information together with the global sequence of the goals' visits to find a path improving the autonomous navigation. Experimental results in real outdoor and indoor environments indicate that the framework provides paths that effectively decreases the localization uncertainty; thus, increases the reliability of the autonomous goals' visits.

In this paper, we discuss the problem of goal assignment in the multi-robot exploration task. The presented work is focused on the underlying optimal assignment problem of the multi-robot task allocation that is addressed by three state-of-the art approaches. In addition, we propose a novel exploration strategy considering allocation of all current goals (not only immediate goal) for each robot, which leads to the multiple traveling salesman problem formulation. Although the problem is strongly NP-hard, we show its approximate solution is computationally feasible and its overall requirements are competitive to the previous approaches. The proposed approach and three well-known approaches are compared in series of problems considering various numbers of robots and sensor ranges. Based on the evaluation of the results the proposed exploration strategy provides shorter exploration times than the former approaches.

Several navigation tasks utilizing a low-cost Micro Aerial Vehicle (MAV) platform AR-drone are presented in this paper to show how it can be used in an experimental verification of scientific theories and developed methodologies. An important part of this paper is an attached video showing a set of such experiments. The presented methods rely on visual navigation and localization using on-board cameras of the AR-drone employed in the control feedback. The aim of this paper is to demonstrate flight performance of this platform in real world scenarios of mobile robotics.

A navigation and stabilization scheme for 3D heterogeneous (UAVs and UGVs) formations acting under a hawk-eye like relative localization is presented in this paper. We formulate a novel Model Predictive Control (MPC) based concept for formation driving in a leader-follower constellation into a required target region. The formation to target region problem in 3D is solved using the MPC methodology for both: i) the trajectory planning and control of a virtual leader, and ii) the control and stabilization of followers - UAVs and UGVs. The core of the method lies in a novel avoidance function based on a model of the formation respecting requirements of the direct visibility between the team members in environment with obstacles, which is crucial for the hawk-eye localization.

Performance of exploration strategies strongly depends on the process of determination of a next robot goal. Current approaches define different utility functions how to evaluate and select possible next goal candidates. One of the mostly used evaluation criteria is the distance cost that prefers candidates close to the current robot position. If this is the only criterion, simply the nearest candidate is chosen as the next goal. Although this criterion is simple to implement and gives feasible results there are situations where the criterion leads to wrong decisions. This paper presents the distance cost that reflects traveling through all goal candidates. The cost is determined as a solution of the Traveling Salesman Problem using the Chained Lin-Kernighan heuristic. The cost can be used as a stand-alone criterion as well as it can be integrated into complex decision systems. Experimental results for open-space and office-like experiments show that the proposed approach outperforms the standard one in the length of the traversed trajectory during the exploration while the computational burden is not significantly increased.

A system for autonomous roads sweeping by applying formations of mobile robots is presented in this paper. The proposed approach based on Receding Horizon Control solves the formation navigation, planning and stabilization in real-word environments with static and dynamic obstacles. The formations employed for sweeping are built up ad-hoc, taking into account length of robots' effectors (e.g. shovels, sweepers) and width of the working area. Presented method enables to smoothly merge smaller teams with the view of sweeping the larger roads (e.g. runways, highways). The formations can operate in two modes: sweeping and moving. In the sweeping mode, the formations are guided with an aim to effectively cover the cleaning roads, while in the moving mode, the planning system emphasizes the effort to reach a desired target. Furthermore, the moving mode enables to autonomously design complex formation maneuvers, as is reverse driving or turning on spot.

In this paper, we present a framework that applies multiple groups of autonomous snowploughs for efficiently removing the snow from airfields. The proposed approach includes formation stabilization into variable shapes depending on the width of runways. The paper is focused on trajectory planning and control during splitting and coupling of formations for cleaning smaller auxiliary roads surrounding main runways. We propose a general method using a receding horizon control for iterative formation assignment. The algorithm is adapted for the kinematics of car-like robots and can be utilized in arbitrary static and dynamic airport assemblage. The proposed approach has been verified by simulations and by hardware experiments.

This paper describes a design, development and experience with an universal multi-robotic robot control system. The overall message-driven system concept is introduced together with its open modular architecture. Besides the description of functionalities of basic modules, the approach to the system configuration and hardware abstraction layer is discussed. The control system allows to manage different types of robots in one team, whereas each team member is able to share and interchange any information with other team members. Since a lot of robotic tasks are computationally intensive and cannot be computed onboard, the system allows to distribute selected subtasks among remote computers. The system has been designed as a test-bed for research of new algorithms from an area of cooperative robotics.

Robot path planning problem is one of most important task mobile robots. This paper proposes an original approach using a path description by string of cubic splines. Such path is easy executable and natural for car-like robot. Furthermore, it is possible to ensure smooth derivation in connections of particular splines. In this case, the path planning is equivalent to optimization of parameters of splines. An evolutionary technique called particle swarm optimization (PSO) was used hereunder due to its relatively fast convergence and global search character. Various settings of PSO parameters were tested and the best setting was compared to two classical mobile robot path planning algorithms.

Accurate models of the environment are a crucial requirement for autonomous mobile robots. The process of how to acquire knowledge about the operating environment is one of the most challenging problems in this research area. The quality of the model depends on the number and types of sensors used. Occupancy grids are the most common low-level models of the environment used in robotics for fusion of noisy data. This paper first introduces a novel method for building an occupancy grid from a monocular color camera. The next part of the work describes a method for fusion of camera data with data from a rangefinder. The final part presents a new method for measuring the quality of the occupancy grid based on the quality of the path created by the grid. The methods were experimentally verified with an indoor experimental robot at the Czech Technical University.