Accurate robotic external force estimation is fundamental to sensorless physical human-robot interaction (pHRI), as it enables robots to interact with environments compliantly and safely. While the Kalman filter-based generalized momentum force estimation method (KF-GM) is widely adopted, its static covariance matrices and Gaussian noise assumption constrain adaptability and robustness, degrading estimation accuracy. This paper proposes a novel Hybrid Adaptive-Robust Kalman Filtering Approach (HARKF) integrating adaptive Kalman filter (AKF) and robust Kalman filter (RKF), with real-time covariance adjustment and outlier rejection, substantially improving adaptability and robustness. However, the fusing of AKF and RKF introduces inherent inter-filter coupling interferences, significantly compromising estimation accuracy due to incompatible noise adaptation mechanisms. Therefore, a noise-type-based module decoupling scheme and a parameter transfer mechanism are proposed, establishing synergistic collaboration between AKF and RKF, where their complementary mechanisms enable reciprocal reinforcement. The decoupling scheme eliminates cross-coupling through noise characteristic analysis, thus preserving system adaptability while enhancing disturbance robustness, resulting in significantly enhanced force estimation accuracy. The transfer mechanism resolves inter-filter parameter conflicts, thereby considerably improving filtering continuity and estimation robustness. Experimental results indicate that compared with existing Kalman filter-based methods, HARKF exhibits superior force estimation accuracy across diverse interactive scenarios.

We have implemented a practical and high-performance non-removable and non-adjustable online 3D box packing algorithm. The problem to be solved by the algorithm belongs to a type of online 3D box packing problem (3D-BPP), but unlike the traditional 3D box packing problem, only a limited number of boxes to be loaded can be known at a time, so the size of boxed is random for algorithm. The problem also requires that the boxes can't be placed in the buffer or the state of the already loaded boxes can't be changed during the whole process. Due to realistic factors, the packing strategy must also satisfy geometric, stability and orientation constraints. We propose a reward function based on spatial residual value assisting the best deep reinforcement learning algorithm we know right now to solve such a question. The residual value of space means the value of the space that can be used in the future. The algorithm adjusts the network parameters in the Actor-Critic framework based on the impact of the intelligence's strategy on the spatial residual value. Compared with recent online 3D box packing strategies, our algorithm performs better than the best algorithm we know with normal reward function (of course better than all current heuristic methods), better than those learning-based methods (about 9% for space utilization), and have fewer learning iterations to converge (about 1000 in all 8000 episodes).

In this article, a hybrid visual-ranging servoing method is proposed to realize high-precision positioning tasks with a 6-degree of freedom (DOF) manipulator. This method utilizes the image and measurement features directly in the control loop. Without the need of complex image feature design and attitude estimation, this method realizes the 6-DOF control of a robot. A vital challenge in traditional vision-based systems is avoiding local minima and singularity problems. To tackle this issue, a full-rank interaction matrix hybrid visual servo (FRHVS) design criterion is proposed, which guarantees that the hybrid interaction matrix and its pseudoinverse matrix are both full rank. Moreover, the interaction matrix for these hybrid strategies, which combines image features with other sensors features, is derived in an analytical form. Experiments on a 6-DOF manipulator show that the proposed method is effective and has global asymptotic stability and high precision.

Multi-target tracking application for fast-moving workpieces has drawn increasing attention in the industrial field. For the dense, fast moving workpieces with few texture features, traditional cameras get poor quality images with dynamic blur and object adhesion, which makes the detection and tracking of workpieces unreliable. However, the event camera outputs events asynchronously at a microsecond speed when the pixel intensity changes, which can capture the contours of fast-moving workpieces well. In this paper, we propose a parallel two-pipe multi-target tracking algorithm based on the event camera for fast-moving workpieces. RGB-E image obtained by fusing the RGB image and the event solves the unreliable detection caused by dynamic blur and object adhesion. The parallel mechanism ensures that the low-speed detection pipeline does not have much impact on the speed of the high-speed tracking pipeline. Hungarian algorithm is used to associate the detection results obtained by the YOLOv4-tiny detector with the tracking results obtained by the KCF tracker. A correction algorithm based on pixel speed is proposed to synchronize detection results and tracking results. Experimental results prove the proposed algorithm can achieve reliable detection and tracking performance for fast-moving workpieces.

Additive manufacturing technology is widely employed in different fields. The robotic 3D printer is increasingly popular in recent years because it can provide more degree of freedom to improve the printing performance. Our paper illustrates a complete approach for kinematic control of the end- effector trajectory of robotic 3D printer. The forward kinematics and the differentiate kinematics is analyzed to establish the kinematic model of the robot. Then inverse-kinematic control scheme is applied to control the robot following the desired path points. More importantly, a trajectory generation algorithm is proposed to generate the desired trajectory. The simulation results show the validation of the approach.

In the image based visual servo, image Jacobian is vital to the system performance because it is the bridge that warps the velocity in feature space to camera velocity in Cartesian space. However, image Jacobian is sensitive to the camera intrinsic parameters, while the camera intrinsic parameter calibration error is almost unavoidable, which heavily affects the image Jacobian and visual servo process. In this paper, we find the camera intrinsic parameter invariance of the image Jacobian in our 4 DOF visual servo system. The image Jacobian of some geometry features is invariant to the camera intrinsic parameters, indicating that the disturbance in camera intrinsic parameters will not affect the convergence trajectory of those features. To further analyse camera intrinsic invariance, we proposed camera intrinsic parameter Jacobian of Image Jacobian, which can fully describe the camera intrinsic parameter invariance of the image Jacobian. The work in this paper can be used to analyse the system sensitivity to the camera intrinsic parameters. The camera intrinsic invariance is also significant for choosing the visual servo features when designing the visual servo system.

This paper presents an identification method which is based on genetic algorithm (GA) and its improved method to estimate dynamic parameters of industrial robots without load. The procedure consists of the following steps: 1) derivation of the linear form of the dynamic model of the robot according to the Lagrange equation; 2) designing of the excitation trajectory in the form of fifth order Fourier series as exciting trajectory; 3) identification, where genetic algorithm is used to find the global optimal parameters through the genetic exchange between the groups and the survival of the fittest mechanism with the minimum variance between the theoretical torque and the actual torque as the optimization criteria; 4) model validation; 5) analysis of the factors influencing the accuracy of the results in the identification process; 6) proposal of improved method. The experimental results show that the predicted torque and the measured torque obtained by the identification algorithm have a high matching degree, and the model can reflect the actual dynamic characteristics of the robot.