The number of images taken from different view-points plays an important role in the accuracy of object detection in cluttered scenarios. Considering Electric Vehicles battery pack disassembly task, which is a highly dense clutter, a framework is proposed to approximate the number of viewpoints to acquire the accuracy of 3.82 mm in screw detection. The paper explores two methods of image acquisition i) taking multiple 3D images, and ii) taking one 3D image and multiple 2D images and combines it with two methods of image processing i) marking the exact visible center of the screw for calculation, and ii) using a bounding box to mark the screw and calculating the center by intersecting diagonals. The results show that using eight viewpoints with a reliable 3D camera delivers the optimal results.

Multi-Agent Reinforcement Learning (MARL) algorithms based on the Centralised Training Decentralised Execution (CTDE) approach have seen a great deal of interest in recent years. Most of the recent works focus on a specific class of environments built on the StartCraft Multi-Agent Challenge (SMAC) environment suite. However, experiments with the PettingZoo multi-particle environments show poor performance in an interesting subset of tasks. In this paper, the nature of these environments and the reward structures are analyzed. It shows that poor performance in these tasks is not due to a lack of representation power in the individual Q function or mixing functions, but rather a result of convergence to a suboptimal equilibrium of dual channel rewards and the issue of agent-reward decoupling that can be a common problem for many MARL environments. We present reward curriculum-based versions of QMIX and VDN as a solution to these problems and compare their results with the standard algorithms. The results show a clear performance gain in terms of a common cumulative reward metric.

In this paper the developed method for off-line compensation of tool deflections when milling aluminum with an industrial robot is presented. The efficiency of this approach is verified with high precision measurements of deflections using a laser tracker. The compensation method includes both the static milling process model which can predict the mean value components of the tool forces and a new combined local/global approach for estimating the combined stiffnesses of joints. With a process model such as the one presented in this paper and estimates of the robot's joint stiffness values, the tool path can be adjusted to counteract deflections of the tool during milling operations. The model was estimated from a large set of machining experiments and is valid for a given tool and material. The stiffness of each joint is a combination of several effects: (i) stiffness of the links, (ii) stiffness of joint bearings and gears and (iii) stiffness of the position control loops given by the individual axis controller gains in the controller software. The method presented in this paper increases accuracy when using an industrial robot for milling in relatively hard materials.

In this paper a new step-wise approach to kinematic calibration of a 5-DOF Gantry-Tau parallel kinematic machine (PKM) is presented. The approach can be adapted to the modular design of the PKM and the calibration could easily perform part of the assembly instructions for the machine. By using measurements from a laser tracker and least-squares estimates of polynomial functions, a typical accuracy of about 20 micrometer was achieved for the base actuators. The remaining set of 30 general parameters for the hexapod link structure and spherical joint connections were successfully estimated using the Complex search-based evolutionary algorithm.