Vertical farms are to be considered the future of agriculture given that they not only use space and resources efficiently but can also consistently produce large yields. Recently, artificial intelligence has been introduced for use in vertical farms to boost crop yields, and crop growth monitoring is an essential example of the type of automation necessary to manage a vertical farm system. Region of interest predictions are generally used to find crop regions from the color images captured by a camera for the monitoring of growth. However, most deep learning-based prediction approaches are associated with performance degradation issues in the event of high crop densities or when different types of crops are grown together. To address this problem, we introduce a novel method, termed pseudo crop mixing, a model training strategy that targets vertical farms. With a small amount of labeled crop data, the proposed method can achieve optimal performance. This is particularly advantageous for crops with a long growth period, and it also reduces the cost of constructing a dataset that must be frequently updated to support the various crops in existing systems. Additionally, the proposed method demonstrates robustness with new data that were not introduced during the learning process. This advantage can be used for vertical farms that can be efficiently installed and operated in a variety of environments, and because no transfer learning was required, the construction time for container-type vertical farms can be reduced. In experiments, we show that the proposed model achieved a performance of 76.9%, which is 12.5% better than the existing method with a dataset obtained from a container-type indoor vertical farm. Our codes and dataset will be available publicly.

In 3D apple detection, the challenge of direction for apple stem harvesting for agricultural robotics has not yet been resolved. Addressing the issue of determining the stem direction of apples is essential for the harvesting processes employed by automated robots. This research proposes a 3D apple detection framework to identify stem direction. First, we constructed a dataset for 3D apple detection that considers the 3-axis rotation of apples based on stem direction. Secondly, we designed a 3D detection algorithm that not only recognizes the dimensions and location of apples, as existing methods do, but also predicts their 3-axis rotation. Furthermore, we effectively fused 3D point clouds with 2D images to leverage the geometric data from point clouds and the semantic information from images, enhancing the apple detection performance. Experimental results indicated that our method achieved AP@0.25 89.56% for 3D detection by considering apple rotation, surpassing the existing methods. Moreover, we experimentally validated that the proposed loss function most effectively estimated the rotation among the various approaches we explored. This study shows the effectiveness of 3D apple detection with consideration of rotation, emphasizing its potential for practical application in autonomous robotic systems.

We address the problem of 6D object pose estimation in aerosol environments, where RGB and depth sensors experience correlated degradation due to scattering and absorption. Handling such spatially varying degradation typically requires depth restoration, but obtaining ground-truth complete depth in aerosol conditions is prohibitively expensive. To overcome this limitation without relying on costly depth completion, we propose RA6D, a framework that integrates attention-guided reliability modeling with feature distillation. The attention map generated during RGB dehazing reflects aerosol distribution and provides a compact indicator of depth reliability. By embedding this attention as an additional feature in an Attention-Guided Point cloud (AGP), the network can adaptively respond to spatially varying degradation. In addition, to address the scarcity of aerosol-domain data, we employ clean-to-aerosol feature distillation, transferring robust representations learned under clean conditions. Experiments on aerosol benchmarks show that RA6D achieves higher accuracy and significantly faster inference than restoration-based pipelines, offering a practical solution for real-time robotic perception under severe visual degradation.

For autonomous systems operating in degraded visual environments, robust 3D perception depends on accurate, metric-scale depth prediction. Yet, thermal depth completion—despite being well-suited for such conditions—remains relatively underexplored. Conventional depth completion approaches, which fuse sparse LiDAR with RGB images, perform well under well-lit conditions but break down in low light or adverse conditions due to the inherent limitations of RGB sensing. We present the first diffusion-based zero-shot thermal depth completion framework, extending the capability of an RGB-trained foundation model to the thermal domain. Our approach is twofold: First, to bridge the domain gap, we introduce a Thermal-Aware Multispectral Adaptation (TAMA) that leverages paired RGB-thermal images to enhance the model’s understanding of thermal image. Second, we replace the iterative optimization process with our Thermal-aware Proxy Alignment (TAPA) module, which directly integrates thermal guidance and sparse metric constraints into the diffusion sampling process. Our framework enables reliable, metric-scale depth prediction across diverse thermal scenes. Notably, our zero-shot method demonstrates performance competitive with fully-supervised baselines that require task-specific training, establishing a foundation for robust 3D perception in challenging environments.