Object detection, particularly open-vocabulary object detection, plays a crucial role in Earth sciences, such as environmental monitoring, natural disaster assessment, and land-use planning. However, existing open-vocabulary detectors, primarily trained on natural-world images, struggle to generalize to remote sensing images due to a significant data domain gap. Thus, this paper aims to advance the development of open-vocabulary object detection in remote sensing community. To achieve this, we first reformulate the task as Locate Anything on Earth (LAE) with the goal of detecting any novel concepts on Earth. We then developed the LAE-Label Engine which collects, auto-annotates, and unifies up to 10 remote sensing datasets creating the LAE-1M — the first large-scale remote sensing object detection dataset with broad category coverage. Using the LAE-1M, we further propose and train the novel LAE-DINO Model, the first open-vocabulary foundation object detector for the LAE task, featuring Dynamic Vocabulary Construction (DVC) and Visual-Guided Text Prompt Learning (VisGT) modules. DVC dynamically constructs vocabulary for each training batch, while VisGT maps visual features to semantic space, enhancing text features. We comprehensively conduct experiments on established remote sensing benchmark DIOR, DOTAv2.0, as well as our newly introduced 80-class LAE-80C benchmark. Results demonstrate the advantages of the LAE-1M dataset and the effectiveness of the LAE-DINO method.

Human-machine interaction technologies are crucial for enhancing human capabilities in the digital world. The hand, as a primary interaction tool, conveys information through gestures and tactile signals. With advancements in flexible electronics and fuzzy systems, it is now possible to achieve interpretable and accurate interactions using hand gestures and tactile information. This paper introduces a fuzzy-encoded, dual-modal soft glove applied to gesture and grasping object classification. The glove, featuring 10 soft pressure sensors and 6 soft bending sensors, is compactly designed and exhibits excellent sensitivity, with average sensitivities of 133 N$^{-1}$for soft pressure sensors and 13deg$^{-1}$for soft bending sensors. The Fuzzy Mamba Encoding (FuME) classification algorithm, inspired by the interpretability and uncertainty resilience of fuzzy systems and the fitting ability of state-space models, was developed. To the best of our knowledge, this work is the first to combine the Mamba structure with fuzzy intelligent systems. Applied to gesture and grasping object classifications, the glove achieved average accuracies of 96.3% and 94.9%, respectively. The fuzzy encoding leverages membership vectors provided by experts to enhance interpretability. These results demonstrate the glove's effectiveness in signal recording and processing and highlight the powerful synergy between flexible electronic technologies and fuzzy systems.

With the development of collaborative robots, the demand for efficient and safe physical human-robot interaction (pHRI) is significantly increasing. In this paper, a two-loop pHRI controller is proposed to reduce the disagreement in human-robot cooperation and to enhance the level of robot assistance. In the outer loop, a human motion intention estimator is designed, combining the strength of model-free and model-based approaches. It estimates the human's desired movement position and provides the confidence level for the estimated value. Subsequently, the estimated value is tracked by the inner loop controller, and the confidence level is used to adjust the robot's behavior in order to reduce the disagreement. In the inner loop, a neuro-adaptive controller with a variable reference model is designed to achieve the efficient pHRI. A neural network is applied to compensate for the nonlinearity of the robot dynamics, gradually aligning the input-output characteristic of the robot dynamic model with the one of the reference model. To minimize the human-robot disagreement during the collaboration process and to enhance the robot's assistance level, a reinforcement learning method is proposed to adjust parameters of the reference model. The proposed control scheme is implemented on a Franka Panda robot and validated through the point-to-point movement simulation and a real-world human-robot lifting experiment. Results suggest that compared to other methods, the proposed approach can indeed reduce the human-robot disagreement and improve the robot assistance level.