Deep learning-based methods have shown excellent potential on object detection and pose estimation with vast amounts of training data to achieve good performance. Obtaining enough comprehensive manual labeling training data is a time-consuming and error-prone task in industrial scenes, and most current time-saving synthetic data generation methods require detailed high-quality CAD models. Instead, we introduce an image-to-image translation based synthetic data generation approach, which requires only untextured CAD models and a small number of real images. The proposed approach starts with edge maps extracted by mesh model projection. And each edge map is provided with realistic appearance learned from real images through image-to-image translation to achieve patch-level realism, and then superimposed on randomly selected backgrounds with a simple Cut-and-Paste strategy. Experiments on our custom dataset and T-LESS dataset show that our synthetic data is competitive with uncomprehensive real data and gives a considerable improvement when compared to the same degree of model rendering images. In addition, the proposed approach allows for extensibility to accommodate new parts with similar texture and structure. And it shows the potential to train image translation model on only one single object and synthesize data for multiples in one take.

Objective: Paraspinal muscle segmentation and reconstruction from MR images are critical to implement quantitative assessment of chronic and recurrent low back pains. Due to unclear muscle boundaries and shape variations, current segmentation methods demonstrate suboptimal performance with insufficient training samples. This work proposes a novel approach to modeling and inferring muscle shapes that enhances segmentation accuracy and efficiency with few training data. Methods: Firstly, a probabilistic shape model (PSM) based on Fourier basis functions and Gaussian processes (GPs) is designed to encode 3D muscle shapes, where anatomical meanings are attributed to the model's geometric parameters. Muscle shape variations and correlations are described by the GPs of the geometric parameters, which allow a small size of parameters to model the distribution of muscle shapes. Secondly, a Bayesian framework is developed to achieve entire muscle segmentation by posterior estimations. The framework fuses the geometric prior of the PSM with observations of deep-learning-based edge detections (DED) and sparse manual annotations, by which issues of unclear boundaries and shape variations can be compensated. Results and conclusion: Experiments on public and clinical datasets demonstrate that, with just three manually annotated slices, our method achieves a Dice similarity coefficient exceeding 90%, which outperforms other methods. Meanwhile, our method needs only a small training dataset and offers rapid inference speeds in clinical applications. Significance: Our study enables precise assessment of paraspinal muscles in 2D and 3D, aiding clinicians and researchers in understanding muscle changes in various conditions, potentially enhancing treatment outcomes.

Biological fish exhibit a remarkably broad-spectrum visual perception capability. Inspired by the eye arrangement of biological fish, we design a fish-like binocular vision system, thereby endowing underwater bionic robots with an exceptionally broad visual perception capacity. Firstly, based on the design principles of binocular visual field overlap and tangency to streamlined shapes, a fish-like vision system is developed for underwater robots, enabling wide-field underwater perception without a waterproof cover. Secondly, addressing the significant distortion and parallax of the vision system, a visual field stitching algorithm is proposed to merge the binocular fields of view and obtain a complete perception image. Thirdly, an orientation alignment method is proposed that draws scales for yaw and pitch angles in the stitched images to provide a reference for the orientation of objects of interest within the field of view. Finally, underwater experiments evaluate the perception capabilities of the fish-like vision system, confirming the effectiveness of the visual field stitching algorithm and the orientation alignment method. The results show that the constructed vision system, when used underwater, achieves a horizontal field of view of 306.56°. The conducted work advances the visual perception capabilities of underwater robots and presents a novel approach to and insight for fish-inspired visual systems.

Low back pain often involves paraspinal muscle degeneration. Rehabilitation robots can provide high-consistency exercise therapy, but current technologies face the challenge of delivering personalized training to patients with various muscular conditions. To solve the issue, this work proposes an automated framework that translates sparse clinical magnetic resonance imaging (MRI) scans into personalized robotic lumbar rehabilitation strategies. First, a Bayesian fusion method with adaptive observation confidence is proposed to enable automatic posterior inference of the 3D paraspinal muscle geometry from sparse MRI. Key biomechanical parameters, including physiological cross-sectional area and fat infiltration, are extracted from the reconstructed muscle shapes to create a patientspecific musculoskeletal model. Based on the personalized model, a hierarchical optimization framework is developed to generate rehabilitation strategies of the combined multi-degree-of-freedom (multi-DOF) motions and dynamic interaction forces to maximize the target muscle activation. Validation on multi-center datasets demonstrates 90% dice similarity for the muscle reconstruction. Personalized validation experiments on volunteers with varying muscle fat infiltration levels revealed that conventional empirical force strategies failed to adapt to individual differences, leading to risks of activation overload or insufficient stimulation. In contrast, the proposed personalized strategy reduced the activation level variance by 60.49% compared to the empirical strategy and maintained the target activation error within 4%. The results demonstrate that the proposed framework significantly mitigates individual uncertainties, ensuring both safety and effectiveness in robotic rehabilitation.