Object detection typically assumes that training and test data are drawn from an identical distribution, which, however, does not always hold in practice. Such a distribution mismatch will lead to a significant performance drop. In this work, we aim to improve the cross-domain robustness of object detection. We tackle the domain shift on two levels: 1) the image-level shift, such as image style, illumination, etc., and 2) the instance-level shift, such as object appearance, size, etc. We build our approach based on the recent state-of-the-art Faster R-CNN model, and design two domain adaptation components, on image level and instance level, to reduce the domain discrepancy. The two domain adaptation components are based on H-divergence theory, and are implemented by learning a domain classifier in adversarial training manner. The domain classifiers on different levels are further reinforced with a consistency regularization to learn a domain-invariant region proposal network (RPN) in the Faster R-CNN model. We evaluate our newly proposed approach using multiple datasets including Cityscapes, KITTI, SIM10K, etc. The results demonstrate the effectiveness of our proposed approach for robust object detection in various domain shift scenarios.

Although the functional properties of shark skin have been of considerable interest to both biologists and engineers because of the complex hydrodynamic effects of surface roughness, no study to date has successfully fabricated a flexible biomimetic shark skin that allows detailed study of hydrodynamic function. We present the first study of the design, fabrication and hydrodynamic testing of a synthetic, flexible, shark skin membrane. A three-dimensional (3D) model of shark skin denticles was constructed using micro-CT imaging of the skin of the shortfin mako (Isurus oxyrinchus). Using 3D printing, thousands of rigid synthetic shark denticles were placed on flexible membranes in a controlled, linear-arrayed pattern. This flexible 3D printed shark skin model was then tested in water using a robotic flapping device that allowed us to either hold the models in a stationary position or move them dynamically at their self-propelled swimming speed. Compared with a smooth control model without denticles, the 3D printed shark skin showed increased swimming speed with reduced energy consumption under certain motion programs. For example, at a heave frequency of 1.5 Hz and an amplitude of ±1 cm, swimming speed increased by 6.6% and the energy cost-of-transport was reduced by 5.9%. In addition, a leadingedge vortex with greater vorticity than the smooth control was generated by the 3D printed shark skin, which may explain the increased swimming speeds. The ability to fabricate synthetic biomimetic shark skin opens up a wide array of possible manipulations of surface roughness parameters, and the ability to examine the hydrodynamic consequences of diverse skin denticle shapes present in different shark species.

Exploiting synthetic data to learn deep models has attracted increasing attention in recent years. However, the intrinsic domain difference between synthetic and real images usually causes a significant performance drop when applying the learned model to real world scenarios. This is mainly due to two reasons: 1) the model overfits to synthetic images, making the convolutional filters incompetent to extract informative representation for real images; 2) there is a distribution difference between synthetic and real data, which is also known as the domain adaptation problem. To this end, we propose a new reality oriented adaptation approach for urban scene semantic segmentation by learning from synthetic data. First, we propose a target guided distillation approach to learn the real image style, which is achieved by training the segmentation model to imitate a pretrained real style model using real images. Second, we further take advantage of the intrinsic spatial structure presented in urban scene images, and propose a spatialaware adaptation scheme to effectively align the distribution of two domains. These two modules can be readily integrated with existing state-of-the-art semantic segmentation networks to improve their generalizability when adapting from synthetic to real urban scenes. We evaluate the proposed method on Cityscapes dataset by adapting from GTAV and SYNTHIA datasets, where the results demonstrate the effectiveness of our method.• real distribution orientation: To deal with the dis-

Remoras of the ray-finned fish family Echeneidae have the remarkable ability to attach to diverse marine animals using a highly modified dorsal fin that forms an adhesive disc, which enables hitchhiking on fast-swimming hosts despite high magnitudes of fluid shear. We present the design of a biologically analogous, multimaterial biomimetic remora disc based on detailed morphological and kinematic investigations of the slender sharksucker (Echeneis naucrates). We used multimaterial three-dimensional printing techniques to fabricate the main disc structure whose stiffness spans three orders of magnitude. To incorporate structures that mimic the functionality of the remora lamellae, we fabricated carbon fiber spinules (270 m base diameter) using laser machining techniques and attached them to soft actuator-controlled lamellae. Our biomimetic prototype can attach to different surfaces and generate considerable pull-off force-up to 340 times the weight of the disc prototype. The rigid spinules and soft material overlaying the lamellae engage with the surface when rotated, just like the discs of live remoras. The biomimetic kinematics result in significantly enhanced frictional forces across the disc on substrates of different roughness. Using our prototype, we have designed an underwater robot capable of strong adhesion and hitchhiking on a variety of surfaces (including smooth, rough, and compliant surfaces, as well as shark skin). Our results demonstrate that there is promise for the development of high-performance bioinspired robotic systems that may be used in a number of applications based on an understanding of the adhesive mechanisms used by remoras.

Many biological organisms can tune their mechanical properties to adapt to environments in multistable modes, but the current synthetic materials, with bistable states, have a limited ability to alter mechanical stiffness. Here, we constructed programmable organohydrogels with multistable mechanical states by an on-demand modular assembly of noneutectic phase transition components inside microrganogel inclusions. The resultant multiphase organohydrogel exhibits precisely controllable thermo-induced stepwise switching (i.e., triple, quadruple, and quintuple switching) mechanics and a self-healing property. The organohydrogel was introduced into the design of soft-matter machines, yielding a soft gripper with adaptive grasping through stiffness matching with various objects under pneumatic-thermal hybrid actuation. Meanwhile, a programmable adhesion of octopus-inspired robotic tentacles on a wide range of surface morphologies was realized. These results demonstrated the applicability of these organohydrogels in lifelike soft robotics in unconstructed and human body environments.

Liver tumors segmentation from computed tomography (CT) images is an essential task for diagnosis and treatments of liver cancer. However, it is difficult owing to the variability of appearances, fuzzy boundaries, heterogeneous densities, shapes and sizes of lesions. In this paper, an automatic method based on convolutional neural networks (CNNs) is presented to segment lesions from CT images. The CNNs is one of deep learning models with some convolutional filters which can learn hierarchical features from data. We compared the CNNs model to popular machine learning algorithms: AdaBoost, Random Forests (RF), and support vector machine (SVM). These classifiers were trained by handcrafted features containing mean, variance, and contextual features. Experimental evaluation was performed on 30 portal phase enhanced CT images using leave-one-out cross validation. The average Dice Similarity Coefficient (DSC), precision, and recall achieved of 80.06% ± 1.63%, 82.67% ± 1.43%, and 84.34% ± 1.61%, respectively. The results show that the CNNs method has better performance than other methods and is promising in liver tumor segmentation.

No abstract

Octopuses can employ their tapered arms to catch prey of all shapes and sizes due to their dexterity, flexibility, and gripping power. Intrigued by variability in arm taper angle between different octopus species, we explored the utility of designing soft actuators exhibiting a distinctive conical geometry, compared with more traditional cylindrical forms. We find that these octopus-inspired conical-shaped actuators exhibit a wide range of bending curvatures that can be tuned by simply altering their taper angle and they also demonstrate greater flexibility compared with their cylindrical counterparts. The taper angle and bending curvature are inversely related, whereas taper angle and applied bending force are directly related. To further expand the functionality of our soft actuators, we incorporated vacuum-actuated suckers into the actuators for the production of a fully integrated octopus arm-inspired gripper. Notably, our results reveal that because of their enhanced flexibility, these tapered actuators with suckers have better gripping power than their cylindrical-shaped counterparts and require significantly larger forces to be detached from both flat and curved surfaces. Finally, we show that by choosing appropriate taper angles, our tapered actuators with suckers can grip, move, and place a remarkably wide range of objects with flat, nonplanar, smooth, or rough surfaces, as well as retrieve objects through narrow openings. The results from this study not only provide new design insights into the creation of next-generation soft actuators for gripping a wide range of morphologically diverse objects but also contribute to our understanding of the functional significance of arm taper angle variability across octopus species.