In this paper, we address the problem of increasing the precision of dense direct stereo visual odometry methods. Dense methods need a dense depth map to generate warped images (virtual views) that will match with reference images if the estimated pose is good. Previous works have shown that generating the depth map by machine learning methods leads to very good odometry results. However, machine learning methods generate hallucinated depths even in areas where it is impossible to estimate the depth due to several reasons, like occlusions, homogeneous areas, etc. Generally, this produces wrong depth estimation that leads to errors in odometry estimation. To avoid this problem, we propose a new approach to generate multiple masks that will be combined to discard wrong pixels and therefore increase the accuracy of visual odometry. Our key contribution is to use the multiple masks not only in the odometry computation but also to improve the learning of the neural network for depth map generation. Experiments on several datasets show that masked dense direct stereo visual odometry provides much more accurate results than previous approaches in the literature.

This letter deals with a novel flying mechanism. Inspired from parallel manipulators, this flying robot is composed of three quadrotors linked by a rigid articulated architecture composed of three legs and a platform. Associating quadrotor comanipulation and rigid bodies, this new design offers novel possibilities for aerial robotics and manipulation. Previous work leads to the design of a flying parallel robot with two quadrotors. However, this robot did not allow the full control of the six degrees of freedom of its end-effector. With an additional quadrotor, this study seeks to obtain a full control of the platform position and orientation. To prove this property, the kinematic constraints are verified through screw theory. Then, the dynamic model is established and a decoupling property leads to the design of a specific controller for the platform and legs configurations. ADAMS/SIMULINK cosimulations validate the theoretical developments.

Finding solutions to traffic congestion is an active area of research. Many ideas have been proposed to reduce this problem, among of this ideas is moving in platoon. The constant time headway policy (CTH) is a very important platoon control policy, but it is too conservative and induces large inter-vehicle distances. Recently, we have proposed a modification of CTH [1], [2]. This modification reduces inter-vehicle distances and makes CTH very practical. This paper focuses on the control of platoons in urban areas. To control the vehicles, we assume that the longitudinal and the lateral dynamics are decoupled. We take into account a simplified engine model. We linearize the two dynamics using exact linearisation technique. Then, we use the modified CTH control law, adapted to urban platoons, for the longitudinal control and the robust sliding mode control for lateral control. The stability and the safety of the platoon are also studied. The conditions of stability of homogeneous and nonhomogeneous platoons are established. The conditions to verify the safety of the platoon for the longitudinal control (assuming stable and accurate lateral control) are exhibited. The weaknesses (large inter-vehicle distance, weak stability near low frequencies) of the CTH are solved. The improved performance and the safety of the platoon are verified by simulation using TORCS (The Open Racing Car Simulator). A platoon consisting of ten vehicles is created and tested on a curved track, keeping a small desired intervehicle distance. The stability and safety of the longitudinal and lateral controls are tested in many scenarios. These scenarios include platoon creation, changing the speed and emergency stop on straight and curved tracks. The results demonstrate the effectiveness of the proposed approach.

Self-driving car's navigation requires a very precise localization covering wide areas and long distances. Moreover, they have to do it at faster speeds than conventional mobile robots. This paper reports on an efficient technique to optimize the position of a sequence of maps along a journey. We take advantage of the short-term precision and reduced space on disk of the localization using 2D occupancy grid maps, from now on called sub-maps, as well as, the long-term global consistency of a Kalman filter that fuses odometry and GPS measurements. In our approach, horizontal planar LiDARs and odometry measurements are used to perform 2D-SLAM generating the sub-maps, and the EKF to generate the trajectory followed by the car in global coordinates. During the trip, after finishing each sub-map, a relaxation process is applied to a set of the last sub-maps to position them globally using both, global and map's local path. The importance of this method lies on its performance, expending low computing resources, so it can work in real time on a computer with conventional characteristics and on its robustness which makes it suitable for being used on a self-driving car as it doesn't depend excessively on the availability of GPS signal or the eventual appearance of moving objects around the car. Extensive testing has been performed in the suburbs and in the down-town of Nantes (France) covering a distance of 25 kilometers with different traffic conditions obtaining satisfactory results for autonomous driving.

Optimizing the inter-distances between vehicles is very important to reduce traffic congestion on highways. Variable spacing and constant spacing are the two policies for the longitudinal control of platoons. Variable spacing doesn't require a lot of data (position, speed...) from other vehicles, and string stability can be obtained using on-board information only. However, inter-vehicle distances are very large, and hence traffic density is low. Constant spacing offers string stability with high traffic density, but it requires data communication between the vehicles, at least from the leader. In this paper, a new platoon model and a modification of the variable spacing policy are proposed. This modification is effective to decrease the distances between the cars, making them nearly equal to the constant spacing policy. It also enables increasing string stability. This new approach doesn't require heavy communication between the vehicles. The new model is based on an unidirectional spring-damper model between vehicles, with the vehicles loaded on a virtual flatbed tow truck. From this configuration, conditions of stability and safety of a homogeneous platoon are derived. Based on this new model, a control has been derived and evaluated by simulation with a perfect system model using Matlab, and with a more realistic vehicle model using TORCS (The Open Racing Car Simulator). The simulation consists of a platoon of ten vehicles, moving on highways, with a desired inter-vehicle distance equal to 1 meter. The stability and the safety of the platoon are tested during platoon creation, changing the speed and emergency stop. The good results demonstrate the effectiveness of the new approach.

Robot manipulators, as general-purposemachines, can be used to perform various tasks. Though, adaptations to specific scenarios require of some technical efforts. In particular, the descriptions of the task result in a robot program which must be modified whenever changes are introduced. Another source of variations are undesired changes due to the entropic properties of systems; in effect, robots must be re-calibrated with certain frequency to produce the desired results. To ensure adaptability, cognitive robotists aim to design systems capable of learning and decision making. Moreover, control techniques such as visual-servoing allow robust control under inaccuracies in the estimates of the system's parameters. This paper reports the design of a platform called CRR, which combines the computational cognition paradigm for decision making and learning, with the visual-servoing control technique for the automation of manipulative tasks.

Recently in [1], a modification of classical constant time headway policy (CTH) was proposed in order to make CTH very practical and easy to use in real applications. This modification was tested and its benefits were shown only for highways application. In this paper, this modification is generalized in order to make it applicable in urban environment. Dynamic and kinematic models of the vehicle are mixted without accounting wheel slip. By using exact linearization technique, lateral and longitudinal dynamics become decoupled. Stability and accuracy of the global system are checked. Using TORCS [8] in simulation, effectiveness of the proposed modification and its potential effect on traffic density are reported.

In this paper, we present a software platform (SoViN) dedicated to visual memory management and vision-based navigation of autonomous vehicles. This s oftware allows to achieve navigation tasks in large scale environments using natural landmarks.It has especially been designed to pro- totype visual memory-based strategies. Such approaches ha ve the major advantage that only key views and related image descriptors are stored. This processis thus expected to be efficient by means of 1) memory needed to store data and 2) computational c ost. These points are crucial issues for real-time navigation in large scale environment . We will see that SoViN allows to meet these expectations.

Several model based techniques have been used to apply various domestic service tasks on humanoid robots (through teleoperation, learning, ... ). But for many reasons, it is more suitable to study the interaction between the robot and its environment using the Sensor Based Control in these cases. In this paper we present a work of integration of real-time visual servoing techniques in performing self localization and different manipulation tasks on a humanoid robot in closed loop.Real-time model based tracking techniques are used to apply 3D visual servoing tasks on the Nao humanoid robot. Elementary tasks used by the robot to perform a concrete scenario are detailed with their corresponding control laws. Experimental results are presented for the following tasks: self-localization of the robot while walking, head servoing for the visibility task, detection, tracking and manipulation of environment's objects.

can be found at: The International Journal of Robotics Research Additional services and information for http://ijr.sagepub.com/cgi/alerts Email Alerts: http://ijr.sagepub.com/subscriptions Subscriptions: http://www.sagepub.com/journalsReprints.nav Reprints: http://www.sagepub.com/journalsPermissions.nav Permissions: Abstract One of the main drawbacks of vision-based control that remains unsolved is the poor dynamic performances caused by the low acquisition frequency of the vision systems and the time latency due to processing. We propose in this paper to face the challenge of designing a high-performance dynamic visual servo control scheme. Two versatile control laws are developed in this paper: a position-based dynamic visual servoing and an image-based dynamic visual servoing. Both control laws are designed to compute the control torques exclusively from a sequential acquisition of regions of interest containing the visual features to achieve an accurate trajectory tracking. The presented experiments on vision-based dynamic control of a high-speed parallel robot show that the proposed control schemes can perform better than joint-based computed torque control.

Ce papier presente une architecture multirobots permettant une allocation automatique de plusieurs objectifs sur une flotte de robots. Le challenge consiste a rendre des robots autonomes pour realiser cooperativement leur mission sans qu'un plan soit predefini. Cette architecture, appelee PRDC, est basee sur 4 modules (Perception, Representation, Deliberation et Controle). Nous nous interessons plus particulierement au module de deliberation en considerant le probleme des voyageurs de commerce cooperatifs dans un environnement incertain. L'objectif des robots est alors de visiter un ensemble de points d'interet representes dans une carte topologique stochastique (Road-Map). Le processus propose pour la construction des politiques collaboratives est distribue. Chaque robot calcule ses politiques individuelles possibles de facon a negocier collectivement l'allocation des points d'interet entre les membres de la flotte. Enfin, l'approche est evaluee via un important nombre de simulations

In this paper, we present a complete framework for autonomous vehicle navigation using a single camera and natural landmarks. When navigating in an unknown environment for the first time, usual behavior consists of memorizing some key views along the performed path to use these references as checkpoints for future navigation missions. The navigation framework for the wheeled vehicles presented in this paper is based on this assumption. During a human-guided learning step, the vehicle performs paths that are sampled and stored as a set of ordered key images, as acquired by an embedded camera. The visual paths are topologically organized, providing a visual memory of the environment. Given an image of the visual memory as a target, the vehicle navigation mission is defined as a concatenation of visual path subsets called visual routes. When autonomously running, the control guides the vehicle along the reference visual route without explicitly planning any trajectory. The control consists of a vision-based control law that is adapted to the nonholonomic constraint. Our navigation framework has been designed for a generic class of cameras (including conventional, catadioptric, and fisheye cameras). Experiments with an urban electric vehicle navigating in an outdoor environment have been carried out with a fisheye camera along a 750-m-long trajectory. Results validate our approach.

Summary According to their specific geometric and dynamic characteristics (small wheelbase and track, small weight, huge reachable speeds…), All-Terrain Vehicles (ATVs - as quad bikes) are very useful and driveable. These specificities permit to realize extra agricultural tasks (spreading, spraying, displacements…) in an easier way than using once more an heavy farm tractor. Unfortunately, the growing popularity of quad bikes in the agricultural area is accompanied by an increasing number of accidents and particularly lateral rollovers. Therefore, the estimation of hazardous situations is a preliminary step in the design of active security devices dedicated to All-Terrain vehicles (ATVs). This paper proposes a rollover metric dedicated to the lateral hazardous situations estimation. It is based on the computation of the Lateral Load Transfer (LLT) according to a backstepping observer dedicated to grip conditions estimation. Next, the maximum vehicle velocity, compatible with a safe motion over some horizon of prediction, is computed via Predictive Functional Control (PFC), and can then be applied, if needed, to the vehicle actuator to prevent from rollover. Capabilities of the proposed metric and device are demonstrated and discussed via both an advanced simulation testbed (that has proved to supply results very close to experimental ones) and full scale experiments.

In this paper, we present a generic framework for urban vehicle navigation using a topological map. This map i s built by taking into account the non-holonomic behaviour of the vehicle. After a localization step, a sensory route is extra cted to reach a goal. This route is followed using a sensor-based con trol strategy, based on the vehicle model and computed from the state extracted from the current and the desired sensory ima ges. In that aim, a generic model is proposed for visual sensors. Experiments with an urban electric vehicle navigating in an outdoor environment have been carried out with a fisheye came ra using a single camera and natural landmarks. A navigation al ong a 1700-meter-long trajectory validates our approach.

This paper describes a navigation strategy ded- icated to non-holonomic wheeled mobile robot using omnidi- rectional cameras. During an off-line learning step, the robot performs paths which are sampled and stored as a set of ordered key images acquired by an embedded camera. The obtained visual paths are topologically organized and provide a memory of omnidirectional images. Given an image of one of the visual paths as a target, the robot navigation mission is defined as a concatenation of visual path subsets, called visual route. The robot is then controlled by a vision-based control law adapted to omnidirectional cameras and to its nonholonomic constraint to follow the reference visual route. Simulations as well as real experimental results illustrate the validity of the pr esented framework. I. I NTRODUCTION