International audience

In this paper, an adaptive and versatile robotic grasping system is presented that is able to manipulate manufactured objects in production factories with a 2-finger gripper. A pick and place scenario based on deep learning framework is implemented and is achieved based on the following main steps: detection of the manufactured objects in the global scene observed by a first RGB-D camera using a first deep learning network, estimation of the object pose using 2D bounding box coordinates and depth information, motion of the arm above the object in an approach pose using Kinematics and Dynamics Library (KDL), recognition of the object's face using a second deep learning network and information coming from a second RGB-D camera setup on the arm wrist, decision on the optimal grasping mode (opening or closing the fingers), execution of the grasping action. The developed system is validated practically by experiments in real world settings using a mobile manipulator platform consisting of 6 DoF robot arm with a 2-finger gripper setup on a mobile robot equipped by two RGB-D cameras.

Since a few years, the development of robots introduced the Cloud Computing paradigm in robotics field as “Cloud Robotics” generation. This new concept allows robots to outsource computing capabilities over the Cloud, whither the Computing involves extra power requirements that could reduce the duration of the process. In this paper, we outline a new Cloud Robotics architecture for NAO robots. Based on Robot Operating System (ROS) middleware, this approach integrates Web services technologies in a Cloud Computing environment, in order to enable Nao robots to consume their packages as an on-demand solution (as a service).

This paper presents a XY-Theta micrometric platform, which is extremely compact and offers a wide 300 × 300 mm workspace. This platform is held by a serial kinematic chain of four revolute joints, constituting a redundant robot. Each point of the horizontal platform can be positioned under a vertical axis in a two-step approach: in a coarse positioning mode, the four axes are controlled to position and orientate the object with a position error less than 7 µm; in a fine mode, two axes are mechanically blocked while two others are controlled to reduce the final position error below 2 µm. The choice of the blocked and moving axes depends on the lever arm length and the mechanism is designed to optimize the link lengths to reduce the final position error. The aim of the paper is to characterize the platform repeatability performances. An estimation of the repeatability is performed with a camera. These results are then compared to previous results based on the stationary cube method. The two measurements methods lead to similar results with a repeatability close to 2 µm showing a significant improvement of the performances.

The paper aims at modeling and quantifying robot precision when it is possible to obtain information from external sensors in the operating area. The author proves that in this situation, neither the pose repeatability, nor the pose accuracy are adequate to calculate the maximum position error. A new paradigm is then proposed: granulous space modeling which combines spatial resolution and actuators' repeatability. This stochastic modeling is first detailed in unidimensional space then in the case of bidimensional space. The methodology to compute the maximal position error is given and compared with other approaches.

The paper deals with the question of robot precision and how to characterise repeatability. Hence ISO and ANSI repeatability indexes advantages and drawbacks are analysed. A new intrinsic repeatability index is proposed that can estimate the robot endpoint position variability satisfying the non-bias and convergence conditions. Computation of this index is performed using simulated straight and drifting trajectories. Influence of load on repeatability is studied using an experimental determination of an angular covariance matrix. Therefrom intrinsic repeatability can be computed in every workspace location using only this covariance matrix and the stochastic ellipsoid theory.

The paper presents innovative kinematics and control of a planar redundant robot designed to improve spatial resolution by a factor 5. This result is obtained in a restricted area of the workspace using position information from external sensors. This innovation results from a clearer understanding of the factors that influence the robot micrometric behavior: axes control resolution generates a set of attainable points in the robot workspace and different spatial resolution patterns appear when introducing redundancy depending on the final axes chosen to correct the position.

In this paper, a new method for the estimation of orientation repeatability index is proposed for industrial manipulator robots. First, we compute orientation repeatability in different locations of the workspace using the experimental covariance matrix and the stochastic ellipsoid modeling. Then we display experimental results about the direct measurement of orientation repeatability for an industrial Samsung robot in different workspace locations and with different loads. The two proposed procedures are compared. We analyze the incidence of workspace location on orientation repeatability and bring additional results to the existing literature.

The paper proposes a new procedure to compare the repeatability of serial and parallel robots based on the stochastic ellipsoid theory. The ISO9283 position repeatability index is estimated but also other performance criteria built upon the stochastic ellipsoid geometrical characteristics. The choice of the best comparison criterion is investigated and different solutions are proposed, associated with the task specificity. For each criterion, maps are built to determine the set of workspace points where the serial robot is better than the parallel robot. The ratio of the workspace surface where one robot is better than the other is computed and the results are analysed. Contrary to the common opinion that parallel robots are more accurate than serial robots, we prove here that the repeatability performance depends mainly on the chosen performance criterion. Another result found is that the considered R̲RRRR̲ parallel robot keeps the same repeatability in all its workspace.

We study the statistical relationship between angular position and target for industrial robots on a micrometric scale and this leads us to understand the angular position stochastic structure. The concept of granular angular space is introduced and transposed in the Cartesian space. Modeling is based on experimental work performed for a Kuka and a Samsung robot. The influence of workspace location, posture and angular granularity ratio on the Cartesian granular space are then analysed.

In this paper, we display experimental results about the measurement of repeatability of an industrial Kuka robot. We first study the distributions of the angular positions and show that these distributions can be considered as Gaussian. We compute repeatability in different locations of the workspace using the experimental angular covariance matrix and the stochastic ellipsoid modeling. We measure repeatability and observe a high variability. We explain the phenomenon by drawing the distribution of the 30 sample repeatability index. We then compare the computed andmeasured repeatability and conclude that our modeling gives good results. We analyse the incidence of weight and workspace location on repeatability and bring additional results to the existing literature.

Most industrial manipulator robots are used in assembly tasks. Their manufacturers use repeatability parameters to show their effectiveness. IS09283 standard details the process of measuring repeatability. As loads, speed and other various factors affect repeatability, a high number of experimental trials are needed to obtain significant values. Moreover, repeatability is also affected by the location of the robot's endpoint in its workspace. In this paper, we construct a stochastic model to evaluate repeatability in the whole workspace of the robot. For this, we only need to determine the robot geometry and sensor sensitivity. Consequently, it is possible with only a few experimental measures to map robot repeatability and obtain relevant information about the spatial error distribution around the points' mean position. This method has been applied to a SCARA robot and has led to specific mapping. Experimental results have been compared with the results of the stochastic ellipsoid model.