This study presents a method for extracting crystal boundaries from a single macro image for robotic laboratory automation. The method is part of the robotic preparation of crystal samples for physical experiments. Detection and precise shape extraction are challenging due to the diverse optical properties of crystals, including transparency, reflections, and partial visibility of the vacuum-tweezer pipette tip that handles the crystal. To address these challenges, we propose a perception method that prioritizes intensity gradients over absolute pixel values. The method uses the Sobel-Feldman operator for edge detection and utilizes α -shapes to reconstruct non-convex crystal boundaries from the resulting point sets. To handle the tweezer visibility, a sliding-window algebraic circle fitting algorithm identifies and isolates the circular pipette tip, preventing shape distortion. Our approach enables the precise determination of 2D boundaries across various crystal types without requiring depth sensors or prior material knowledge. On a dataset of 20 annotated macro images, the method achieves a mean IoU of 97.29

We present a robot kinematic calibration method that combines complementary calibration approaches: self-contact, planar constraints, and self-observation. We analyze the estimation of the end effector parameters, joint offsets of the manipulators, and calibration of the complete kinematic chain (DH parameters). The results are compared with ground truth measurements provided by a laser tracker. Our main findings are: (1) When applying the complementary calibration approaches in isolation, the self-contact approach yields the best and most stable results. (2) All combinations of more than one approach were always superior to using any single approach in terms of calibration errors and the observability of the estimated parameters. Combining more approaches delivers robot parameters that better generalize to the workspace parts not used for the calibration. (3) Sequential calibration, i.e. calibrating cameras first and then robot kinematics, is more effective than simultaneous calibration of all parameters. In real experiments, we employ two industrial manipulators mounted on a common base. The manipulators are equipped with force/torque sensors at their wrists, with two cameras attached to the robot base, and with special end effectors with fiducial markers. We collect a new comprehensive dataset for robot kinematic calibration and make it publicly available. The dataset and its analysis provide quantitative and qualitative insights that go beyond the specific manipulators used in this work and apply to self-contained robot kinematic calibration in general.

Recognizing pedestrian attributes in camera surveillance images is a very hard problem, due to the lack of high-quality labeled data. In the field of deep learning the semi-supervised learning paradigm provides a possible answer to this problem. We propose a novel semi-supervised model that we call Binary Mean Teacher, tailored for binary classification task of detecting the presence of wearable objects. We train our model in a traditional scenario with a randomly initialized model, but we also explore fine-tuning a model pretrained on a large-scale image dataset. The performance of our model is compared to strong supervised baselines trained or fine-tuned using our dataset and the same augmentation strategy as in our model. We evaluate the impact of various augmentation strategies commonly used in deep learning on the performance of models in our binary classification task. Using only 1000 labeled training images, randomly initialized Binary Mean Teacher model achieves roughly 90% classification accuracy compared to 75% accuracy of randomly initialized supervised model that does not use any augmentations. When both Binary Mean Teacher and the supervised model are pretrained using the ImageNet dataset, and augmentations are used for both models, the Binary Mean Teacher achieves 92% accuracy compared to 90% accuracy of the supervised model.

This work is focused on the use of a CZT detector for a radiation mapping with an industrial robotic arm. Measurements were carried out within the RadioRoSo experiment (Radioactive Waste Robotic Sorter), under the umbrella of EU FP7 project ECHORD++. In tests with a dual-arm robot and standard point sources of 137Cs and 60Co, a Magnox waste was mimicked. Thereafter, for relevant measurement geometries and selected shielding materials, full energy peak efficiencies were calculated using the MCNP transport code. Simulated FEP efficiency values were used for minimum detectable activity assessments for expected measurement conditions. Obtained results would be also applicable for cases of shielded lost/orphan point-like sources.

Artificial neural networks, in particular the deep end-to-end architectures trained by error backpropagation (BP), are currently the topmost used learning systems. However, learning in such systems is only loosely inspired by the actual neural mechanisms. Algorithms based on local activation differences were designed as a biologically plausible alternative to BP. We propose Universal Bidirectional Activation-based Learning, a novel neural model derived from contrastive Hebbian learning. Similarly to what is assumed about learning in the brain, our model defines a single learning rule that can perform multiple ways of learning via special hyperparameters. Unlike others, our model consists of mutually dependent, yet separate weight matrices for different directions of activation propagation. We show that UBAL can learn different tasks (such as pattern retrieval, denoising, or classification) with different setups of the learning hyperparameters. We also demonstrate the performance of our algorithm on a machine learning benchmark (MNIST). The experimental results presented in this paper confirm that UBAL is comparable with a basic version BP-trained multilayer network and the related biologically-motivated models.

Transfer learning is a well known technique to circumvent the problem of small datasets in deep machine learning. It has been successfully used in the field of camera surveillance image processing which suffers from poor data quality and quantity. We focused on the task of wearable object detection, namely distinguishing if a person is or is not wearing a backpack. We created new annotations for the DukeMTMC-attribute dataset to overcome the discrepancies among the attributes. We explored transfer learning with a frozen feature extractor as well as the model fine-tuning, which turned out to perform much better. In both setups we found that the Densenet161 is the best from tested architectures. Our best model achieved about 92% balanced accuracy on the testing set.

The estimation of the fabric material property during the folding is presented. The available techniques for the accurate garment folding rely on known material properties. Currently, the properties are estimated by an operator in advance of folding. We propose an iterative strategy, which updates the property while the garment is folded. The estimation is formulated as an optimisation task. It is based on measurements from a laser range finder. The proposed algorithm improves the estimation iteratively and prevents the garment from slipping at the same time. We demonstrate the estimation procedure for 10 fabric strips of different materials.

Sorting of old and mixed nuclear waste is an essential process in nuclear decommissioning operations. The main bottleneck is manual picking and separation of the materials using remotely operated arms, which is slow and error prone especially with small items. Automation of the process is therefore desirable. In the framework of the newly funded European project ECHORD++, experiment RadioRoSo, a pilot robotic cell is being developed and validated against industrial requirements on a range of sorting tasks. Industrial robots, custom gripper, vision feedback and new manipulation skills will be developed. This paper presents application context, cell layout and sorting approach.

We address the accurate single arm robotic garment folding. The folding capability is influenced mostly by the folding path which is performed by the robotic arm. This paper presents a new method for the folding path generation based on the static equilibrium of forces. The existing approach based on a similar principle confirmed to be accurate for one-dimensional strips only. We generalize the method to two-dimensional shapes by modeling the garment as an elastic shell. The path generated by our method prevents the garment from slipping while folding on a low friction surface. We demonstrate the accuracy of this approach by comparing our paths (a) with the existing method when one-dimensional strips of different materials were modeled, and (b) experimentally with real robotic folding.

We deal with the problem of thin string (1D) or plate (2D) elastic material folding and its modeling. The examples could be metallic wire, metal, kevlar or rubber sheet, fabric, or as in our case, garment. The simplest scenario attempts to fold rectangular sheet in the middle. The quality of the fold is measured by relative displacement of the sheet edges. We use this scenario to analyse the effect of the inaccurate estimation of the material properties on the fold quality. The same method can be used for accurate placing of the elastic sheet in applications, e.g. the industrial production assembly.In our previous work, we designed a model simulating the behavior of homogeneous rectangular garment during a relatively slow folding by a dual-arm robot. The physics based model consists of a set of differential equations derived from the static forces equilibrium. Each folding phase is specified by a set of boundary conditions. The simulation of the garment behavior is computed by solving the boundary value problem. We have shown that the model depends on a single material parameter, which is a weight to stiffness ratio. For a known weight to stiffness ratio, the model is solved numerically to obtain the folding trajectory executed by the robotic arms later.The weight to stiffness ratio can be estimated in the course of folding or manually in advance. The goal of this contribution is to analyse the effect of the ratio inaccurate estimation on the resulting fold. The analysis is performed by simulation and in a real robotic garment folding using the CloPeMa dual-arm robotic testbed. In addition, we consider a situation, in which the weight to stiffness ratio cannot be measured exactly but the range of the ratio values is known. We demonstrate that the fixed value of the ratio produces acceptable fold quality for a reasonable range of the ratio values. We show that only four weight to stiffness ratio values can be used to fold all typical fabrics varying from a soft (e.g. sateen) to a stiff (e.g. denim) material with the reasonable accuracy. Experiments show that for a given range of the weight to stiffness ratio one has to choose the value on the pliable end of the range to achieve acceptable results.

The ability to perform an accurate robotic fold is essential to obtain the properly folded garment. Available solutions rely on a rough folding surface or on a comprehensive simulation, both preventing the garment from slipping on the table during folding. This paper proposes a new algorithm for a folding path design respecting the garment material properties and preventing the garment slipping. The folding path is derived based on the equilibrium of forces under the simplifying assumptions of a rectangular and homogeneous garment. This approach allows folding the rectangular garment on a low friction table surface as we demonstrated in the experiments performed by a dual-arm robotic testbed.

The trajectory performed by a dual-arm robot while folding a piece of garment was studied. The garment folding was improved by adopting here proposed novel circular folding trajectory, which takes the flexibility of the garment into account. The benefit lies in an increased folding precision. In addition, several relaxations of the folding trajectory were introduced, thus enlarging the working space of the dual-arm robot.The new folding trajectory was experimentally verified and compared to the state-of-the-art methods. The advocated approach assumes that the folding trajectory and the robot arms constitute a closed kinematic chain. Closed loop planning techniques introduced recently enable planning of the folding task without solving the inverse kinematics in each planning step. This approach has favourable properties because the model used is extensible. For instance, it is possible to take into account the force/tension applied by the robot grippers to the held garment.

The work addresses the problem of clothing perception and manipulation by a two armed industrial robot aiming at a real-time automated folding of a piece of garment spread out on a flat surface. A complete solution combining vision sensing, garment segmentation and understanding, planning of the manipulation and its real execution on a robot is proposed. A new polygonal model of a garment is introduced. Fitting the model into a segmented garment contour is used to detect garment landmark points. It is shown how folded variants of the unfolded model can be derived automatically. Universality and usefulness of the model is demonstrated by its favorable performance within the whole folding procedure which is applicable to a variety of garments categories (towel, pants, shirt, etc.) and evaluated experimentally using the two armed robot. The principal novelty with respect to the state of the art is in the new garment polygonal model and its manipulation planning algorithm which leads to the speed up by two orders of magnitude.

This report describes how CloPeMa explores visual information and uses it for the scene understanding, planning and manipulation of a garment piece. First, visual sensors implemented at CloPeMa testbed are described. Second, the procedure for grasping a piece of garment is explicated. Third, buttons serving as features are utilized. Fourth, unfolding skills using random forests are described. Fifth, the garment description from a high resolution stereo head is reported. Sixth, a substantial improvement to the state of the art Berkeley garment model is suggested and its experimental validation documented.

Controlling satellite trajectories is an important problem. In [12], an approach to the pole placement for the synthesis of a linear controller has been presented. It leads to solving five polynomial equations in nine unknown elements of the state space matrices of a compensator. This is an underconstrained system and therefore four of the unknown elements need to be considered as free parameters and set to some prior values to obtain a system of five equations in five unknowns. In [12], this system was solved for one chosen set of free parameters by Dixon resultants. In this work, we study and present Gröbner basis solutions to this problem of computation of a dynamic compensator for the satellite for different combinations of free input parameters. We show that the Gröbner basis method for solving systems of polynomial equations leads to very simple solutions for all combinations of free parameters. These solutions require to perform only the Gauss-Jordan elimination of a small matrix and computation of roots of a single variable polynomial. The maximum degree of this polynomial is not greater than six in general but for most combinations of the input free parameters its degree is even lower.

Reference 1] Pavel Krsek, Tomm a s Pajdla, and VV aclav Hlavv a c. Estimation of differential parameters on triangulated surface. Abstract This paper presents an algorithm for estimation of principal curvatures and principal directions from 3D scanned data. The diierential parameters will be base for nding of diierential structures on the surface. This structures should be used for fusion of range data from more views. Input to the algorithm are coordinates of points and the triangulation deening a neighbourhood of points. First, the measured points are locally approximated by an osculating paraboloid. The differential parameters of the paraboloid are considered as an initial estimate of the parameters of the original surface. If the measured points are samples of function z = f (x; y), the paraboloid with axis parallel to z axis can be used for local approximation. The noise usually have large innuence on the initial estimation of parameters. Therefore, the initial estimate is further reened by using estimated parameters from the neighbourhood of each point.

Reference 1] Pavel Krsek, Tomm a s Pajdla, and VV aclav Hlavv a c. Estimation of differential structures on triangulated surfaces. Abstract: This paper presents a method for extracting diierential structures on 3D surfaces and describes an algorithm for estimation of principal curvatures and principal directions from 3D scanned data. The method serves as a base for the extracting intrinsic curves on the surface. The diierential structures are intended for matching partially overlapping range images. The search for a deferential structure which is represented by the curves of innection points is based on estimation of diierential parameters of the surface. Therefore, the reliable estimation of diierential parameters is necessary. The estimation of diierential parameters consists of two steps. First, the measured points are locally approximated by an osculating paraboloid. The diierential parameters of the paraboloid are considered as an initial estimate of the parameters of the original surface. Then, the initial estimate is further reened using estimated parameters from the neighbourhood of each point.

Superimposition is an efficient method for evaluation of coincidence between a skull and a photo portrait. The principle of superimposition method lies in the projection of the skull into the face image. During the projection of an object with a perspective camera, the mapping of a three-dimensional object into a two-dimensional image takes place. The acquired images of the same object are more or less distorted due to various photographic conditions, due to extrinsic and intrinsic parameters of the camera. The distortions have important influence onto reliability of human identification by the superimposition method. Mathematically we can describe most of the distortions. On the basis of the description the divergences could be simulated and in some cases eliminated by geometric transformation of the compared images. We are presenting a mathematical model of the standard projective camera and the mathematical description of distortions which are important for the superimposition process. The results show the distortions and the elimination of the distortions by means of the projection model.

The problem of geometric alignment of two roughly pre-registered, partially overlapping, rigid, noisy 3D point sets is considered. A new natural and simple, robustified extension of the popular Iterative Closest Point (ICP) algorithm [IEEE Trans. Pattern Anal. Machine Intell. 14 (1992) 239] is presented, called Trimmed ICP (TrICP). The new algorithm is based on the consistent use of the Least Trimmed Squares approach in all phases of the operation. Convergence is proved and an efficient implementation is discussed. TrICP is fast, applicable to overlaps under 50%, robust to erroneous and incomplete measurements, and has easy-to-set parameters. ICP is a special case of TrICP when the overlap parameter is 100%. Results of a performance evaluation study on the SQUID database of 1100 shapes are presented. The tests compare TrICP and the Iterative Closest Reciprocal Point algorithm [Fifth International Conference on Computer Vision, 1995].