Inverse kinematics (IK) plays an essential role in the field of robotics, enabling robots to determine the joint configurations required to achieve desired end-effector poses. Gradient-based numerical methods are commonly used for IK problem-solving, where the step size influences the convergence success and solving time. In this work, we propose a novel approach that employs dynamic step sizes based on the relative gradient norm at each iteration, aiming to improve solve rates and overall performance. We evaluated the proposed method on six different kinematic chains of robotic manipulators and compared it with conventional fixed step size approaches. Experimental results demonstrate that the use of dynamic step sizes, specifically the Gradient Scaling method, achieves significant improvements in solve rates, surpassing 150% improvement in various kinematic chains. Moreover, we enhance the solve rate further by incorporating a random restart approach to escape local minima. The proposed approach presents a viable alternative to conventional gradient-based IK methods, automating the step size selection and offering improved efficiency and robustness in solving complex robotic tasks.
This paper describes the development and experimental evaluation of user interfaces of a semi-automatic visual supporting system using a drone for a teleoperated construction robot. While the developed system needs to be operated by the user, the operation of the system should not adversely affect the operation of the construction robot. In this paper, three interfaces are proposed: a head motion interface, a slider interface, and a voice interface. These interfaces are evaluated by an experiment in the teleoperation of the construction robot. As a result of the experiment, we confirmed that the slider interface was an advantage against the others in terms of mental burden and work time.
This paper presents a light-weight convolutional neural network for SEM imaging. Combining nanorobotic manipulation systems with the scanning electron microscope makes precise micro-nano manipulation based on visual feedback become possible. However, the contradiction between imaging speed and resolution hinders the real-time of micro-nano manipulation, which further obstacles to the realization of applying it in mass manufacturing situations. In this paper, we propose a super resolution method called SRSEMNet1 to enhance the imaging quality of SEM by super-resolving low resolution SEM images. In this way, high-quality SEM images could be obtained while consumes less time than acquiring them straightly from SEM, which contributes to the real time of micro-nano manipulation. We analyze the residual learning, skip connections, CBAM, and different loss functions to optimize the network. The method was tested in both well accepted benchmark dataset Urban100 and our own made dataset SSRID for SEM imaging.
Reinforcement learning (RL) has demonstrated potential in addressing deformable object manipulation problems. However, significant challenges remain regarding the complexity of these tasks and the limited efficiency of policy initialization with human demonstrations (HD), which may necessitate substantial data and time for convergence. This paper proposes human demonstrations enhanced deep deterministic policy gradients (HDE-DDPG) to simplify the deformable object manipulation problem and maximize the benefits of HD in RL. The complexity of manipulation tasks is reduced by carrying out conditional policy training and utilizing a fuzzy classification system known as high-dimensional Takagi–Sugeno–Kang (HTSK) for grasp point selection. The output of the HTSK is then fed as input to the actor to determine the placement position. By training the critic to assign higher scores to HD actions during pretraining, the training of the critic is accelerated. As a result of an improved critic, behavior cloning (BC) is used more frequently and appropriately, which fully exploits HD and accelerates the training of the entire system. HDE-DDPG was evaluated through ablation experiments, and the results demonstrate that our method significantly speeds up the agent's training while achieving superior and consistent performance.
Reinforcement learning models have been instrumental in simulating the behaviors of animals and humans, whose actions and choices are largely influenced by rewards. This study proposes a high-level environmental monitoring structure into cognitive models, optimizing the balance between exploration and exploitation in reinforcement learning theory. This approach allows for more flexible adaptation to complex environmental changes. We focus on a multi-armed bandit task, incorporating a hierarchical model that adds a high-level cognition control structure based on the cognitive system. This structure adjusts the balance between exploration and exploitation according to the perceived uncertainty of the environment. In a stable environment, the model minimizes exploration to maximize rewards. In contrast, in an unstable environment, the model promotes rapid exploration to identify the best strategy after environmental changes. This cognition-based hierarchical structure offers valuable insights for future studies about analyzing human responses to environmental uncertainty and exploration behavior.
