Search

Purpose This paper aims to study the application of reinforcement learning (RL) in the control of an output-constrained flapping-wing micro aerial vehicle (FWMAV) with system uncertainty. Design/methodology/approach A six-degrees-of-freedom hummingbird model is used without consideration of the inertial effects of the wings. A RL algorithm based on actor–critic framework is applied, which consists of an actor network with unknown policy gradient and a critic network with unknown value function. Considering the good performance of neural network (NN) in fitting nonlinearity and its optimum characteristics, an actor–critic NN optimization algorithm is designed, in which the actor and critic NNs are used to generate a policy and approximate the cost functions, respectively. In addition, to ensure the safe and stable flight of the FWMAV, a barrier Lyapunov function is used to make the flight states constrained in predefined regions. Based on the Lyapunov stability theory, the stability of the system is analyzed, and finally, the feasibility of RL in the control of a FWMAV is verified through simulation. Findings The proposed RL control scheme works well in ensuring the trajectory tracking of the FWMAV in the presence of output constraint and system uncertainty. Originality/value A novel RL algorithm based on actor–critic framework is applied to the control of a FWMAV with system uncertainty. For the stable and safe flight of the FWMAV, the output constraint problem is considered and solved by barrier Lyapunov function-based control.

As we all know, detection and recognition of distant or relatively small targets is a challenging problem. However, the eagle-eye is known for its sharp eyesight, which is called clairvoyance. Inspired by the structural characteristics and physiological mechanism of eagle-eye, a lightweight vision system is designed for target detection and recognition in this paper. First, a hardware platform is established, which consists of a pan-tilt, a short-focus camera and a long-focus camera. Then, combined with the image processing mechanism of the eagle-eye and deep learning, the cameras with different focal lengths are controlled cooperatively to achieve target searching, edge extraction, target detection and recognition. Finally, by separately detecting the targets with different focal length cameras, it can be found that the confidence of the target detection with the long-focus camera is greatly improved. Experimental results show that the designed lightweight eagle-eye vision system can accurately detect and recognize targets, which has a strong adaptive ability.

Plant biomass are considered as good precursors for synthesizing carbons due to their abundance and non-toxicity. In the synthesis process, many activators are needed to endow carbons with abundant pore structure. However, the traditional activators are highly corrosive and reduce the nitrogen content of carbons, which is not conducive to their applications. Herein, we choose sodium amide (NaNH2), which is less corrosive and strongly nucleophilic, as activator. The research shows that NaNH2 can activate ginkgo leaves at relatively low temperature (500 °C) and obviously improve the structure and composition of their derived carbons. When used as catalysts for the oxidative coupling of benzylamine to imine, the carbon synthesized with the largest amount of NaNH2 activator exhibits the best performance, which can be attributed to the synergistic effects of high surface area, hierarchical structure and abundance active sites. Further, the catalytic performance of carbons derived from apricot leaves and poplar leaves activated by NaNH2 also increases with the increase in the NaNH2 dosage, which indicates that NaNH2 is a widely adaptable activator for plant biomass.

The probe-and-drogue autonomous aerial refueling (AAR) docking suffers the multiwind disturbances leading to the low accuracy of receiver position control and drogue relative position estimation. This article proposes a bionic visual navigation control system in a hardware-in-loop simulation environment and further develops an AAR outfield experiment platform for promoting the success of probe-and-drogue AAR docking. The drogue region and markers are detected by the biological method imitating the eagle-eye color vision mechanism. The different visual navigation methods based on the ellipse fitting, marker matching, and marker prediction are, respectively, applied to the three possible situations in AAR docking for precise pose estimation. Moreover, a relative position control scheme of the receiver, which is constructed by back-stepping design technique and gain-adaptive equivalent sliding mode control, is proposed to compensate multi-wind disturbances and model uncertainties. Fuzzy logic position strategy is designed to compensate swing of the drogue caused by the multiwind disturbances. The AAR outfield experiment platform contains two unmanned aerial vehicles to verify a part of the proposed navigation control method. Extensive experimental results are presented to demonstrate the effectiveness of the bionic visual navigation control system for the high accuracy of the visual navigation and antidisturbance position control.

We synthesized foam-like carbon materials from purslane stem by soxhlet extraction and pyrolysis two steps. The results demonstrate that the solvents used in soxhlet extraction can extract different components from purslane stem and further make significant impacts on the structure and composition of the obtained carbon materials. When applied to catalytic oxidative coupling of amines to imines, the most active catalyst is the carbon material derived from the purslane stem pretreated with ethanol, which can be attributed to its high surface area, large pore size and accessible active sites. In addition, the catalyst shows good stability, which can be recycled at least six times and applicable in the large-scale catalytic reaction. Moreover, the carbon materials derived from celery stem and apricot leaf pretreated with ethanol in soxhlet extraction also show higher catalytic activity than the carbon materials pretreated with ethyl acetate and water, which implies that this simple synthetic process is generally applicable to a certain extent. Hence, this study can provide new way toward the synthesis of high-performance carbon materials from biomass.

Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent overwhelming success of convolutional neural networks (CNNs) for image analysis suggests the value of adapting insight from CNN to the point cloud world. Point clouds inherently lack topological information, so designing a model to recover topology can enrich the representation power of point clouds. To this end, we propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds, including classification and segmentation. EdgeConv acts on graphs dynamically computed in each layer of the network. It is differentiable and can be plugged into existing architectures. Compared to existing modules operating in extrinsic space or treating each point independently, EdgeConv has several appealing properties: It incorporates local neighborhood information; it can be stacked applied to learn global shape properties; and in multi-layer systems affinity in feature space captures semantic characteristics over potentially long distances in the original embedding. We show the performance of our model on standard benchmarks, including ModelNet40, ShapeNetPart, and S3DIS., Army Research Office (Grant W911NF-12-R-0011), Air Force Office of Scientific Research (Award FA9550-19-1-0319), National Science Foundation (Grant IIS-1838071)

Carbon materials with tailorable structures and superior properties have great potential applications in environmental protection, energy conversion, and catalysis. Plant biomass as abundant and green non-toxic raw materials has been considered as good precursors for synthesizing heteroatom-doped carbon materials. However, few studies have been reported on the different natures of carbon materials derived from different parts of the same plant biomass. In this study, we prepared carbon materials from the petioles and blades of apricot leaves by direct pyrolysis without additives. Detailed characterizations indicate that these two carbon materials are similar in element composition and graphitization degree, but differ greatly in surface area and pore volume. These differences can be attributed to the different contents of inorganic salts, vascular bundles, and proteins in petioles and blades. When used as catalysts for the oxidation of ethylbenzene, the petiole-derived carbon shows better catalytic performance than the blades derived carbon due to its high surface area, large average pore size, and doped nitrogen atoms. Furthermore, the carbon catalysts derived from the petioles and blades of poplar leaves and parasol tree leaves show the same difference in catalytic reaction, implying that the above-mentioned conclusion is rather universal, which can provide reference for the synthesis of carbon materials from leaves.

Bionic visual close-range navigation control system and method for the docking stage of probe-and-drogue autonomous aerial refueling (AAR) are developed in this paper. The reasonable bionic visual close-range navigation system is devised to obtain the accurate pose information of drogue which is connected to a flexible hose in the presence of multi-wind disturbances. Correspondingly, the biological eagle-eye-based color detection method is constructed on the basis of the eagle-eye color vision mechanism to detect the drogue region and markers. The different visual navigation methods based on marker matching and ellipse fitting are respectively applied for the two situations: the markers detected normally, and the drogue far away to receiver or the markers blocked partially. Moreover, due to the switch of different control targets and navigation methods (GPS or bionic visual navigation) for controllers in the docking stage, the different relative position precise controllers are designed to control the relative position between the tanker and receiver. The experimental results show that the bionic visual close-range navigation system and method effectively acquire the drogue's pose information in the entire visual docking stage and the designed controllers are suitable to the GPS and bionic visual navigation methods for changing and keeping the relative position between the tanker/receiver or probe/drogue.

We identify the need for an easy-to-use self-driving simulator where game mechanics implicitly encourage high-quality data capture and an associated low-cost physical test platform. We design such a simulator incorporating environmental domain randomization to enhance data generalizability and a low-cost physical test platform running the Robotic Operating System. A toolchain comprising a gamified driving simulator and low-cost vehicle platform is novel and facilitates behavior cloning and domain adaptation without specialized knowledge, supporting crowdsourced data generation. This enables small organizations to develop certain robust and resilient self-driving systems. As proof-of-concept, the simulator is used to capture lane-following data from AI-driven and human-operated agents, with these data training line following Convolutional Neural Networks that transfer without domain adaptation to work on the physical platform.

Fashion recommendation systems are highly desired by customers to find visually-collocated fashion items, such as clothes, shoes, bags, etc. While existing methods demonstrate promising results, they remain lacking in flexibility and diversity, e.g. assuming a fixed number of items or favoring safe but boring recommendations. In this paper, we propose a novel fashion collocation framework, Neural Graph Filtering, that models a flexible set of fashion items via a graph neural network. Specifically, we consider the visual embeddings of each garment as a node in the graph, and describe the inter-garment relationship as the edge between nodes. By applying symmetric operations on the edge vectors, this framework allows varying numbers of inputs/outputs and is invariant to their ordering. We further include a style classifier augmented with focal loss to enable the collocation of significantly diverse styles, which are inherently imbalanced in the training set. To facilitate a comprehensive study on diverse fashion collocation, we reorganize Amazon Fashion dataset with carefully designed evaluation protocols. We evaluate the proposed approach on three popular benchmarks, the Polyvore dataset, the Polyvore-D dataset, and our reorganized Amazon Fashion dataset. Extensive experimental results show that our approach significantly outperforms the state-of-the-art methods with over 10% improvements on the standard AUC metric on the established tasks. More importantly, 82.5% of the users prefer our diverse-style recommendations over other alternatives in a real-world perception study.

Watching sports events via 3D-instead of two-dimensional video streaming allows for increased immersion, e.g. via mixed reality headsets in comparison to traditional screens. So far, capturing 3D video of sports events required expensive outside-in tracking with numerous cameras. This study demonstrates the feasibility of streaming sports content to mixed reality headsets as holographs in real-time using inside-out tracking and low-cost equipment only. We demonstrate our system by streaming a race car on an indoor track as 3D models, which are then rendered in an Magic Leap One headset. An onboard camera, mounted on the race car provides the video stream used to localize the car via computer vision. The localization is estimated by an end-to-end convolutional neural network (CNN). The study compares three state-of-the-art CNN models in their respective accuracy and execution time, with PoseNet + LSTM achieving position and orientation accuracy of 0.35m and 3.95°. The total streaming latency in this study was 1041ms, suggesting technical feasibility of streaming 3D sports content, e.g. on large playgrounds, in near real-time onto mixed-reality headsets.

© 2020 IEEE. Generating 3D point clouds is challenging yet highly desired. This work presents a novel autoregressive model, PointGrow, which can generate diverse and realistic point cloud samples from scratch or conditioned on semantic contexts. This model operates recurrently, with each point sampled according to a conditional distribution given its previously-generated points, allowing inter-point correlations to be well-exploited and 3D shape generative processes to be better interpreted. Since point cloud object shapes are typically encoded by long-range dependencies, we augment our model with dedicated self-attention modules to capture such relations. Extensive evaluations show that PointGrow achieves satisfying performance on both unconditional and conditional point cloud generation tasks, with respect to realism and diversity. Several important applications, such as unsupervised feature learning and shape arithmetic operations, are also demonstrated.

Dynamic modeling and control system design for the hose–drogue system (HDS) in the docking stage of autonomous aerial refueling (AAR) are investigated in this paper. The dynamics and kinematics of hose are modeled via a finite-segment multi-body method, which describes the hose–drogue assembly as a link-connected system. A controllable drogue is connected to the hose for automatically stabilizing the drogue's relative position under the influences of tanker trailing vortex, receiver bow wave, atmospheric turbulence, gust, and wind shear. Thus, a drogue position control law based on fractional-order method is designed to resist the multi-wind disturbances. Noting that it is difficult to tune the parameters of fractional-order controller (FOC), a modified pigeon-inspired optimization (PIO), the hybrid of heterogeneous comprehensive learning strategy and PIO (HCLPIO), is carried out to optimize the parameters of FOC. The simulation results show that the proposed optimized fractional-order feedback controllers effectively stabilize the controllable drogue to swing within an acceptable range.

Ag2Mn5/SiC catalyst is simply obtained by in-situ reaction-induced of their corresponding nitrate compounds onto SiC support, which exhibits excellent catalytic activity and stability in the gas phase oxidation of alcohols (94% conversion of benzyl alcohol and 98% selectivity of benzaldehyde). The catalytic activity could maintain at least 300 h after four regenerations due to the formation of highly active Ag0-Mn3O4 species. XRD and XPS indicate that active Ag0-Mn3O4 species is formed from Ag2O and MnO2 in the reaction conditions. The synergistic effect between 34 nm Ag nanoparticles and 5.5 nm Mn3O4 nanoparticles in alcohol oxidation reaction is prominent, which is testified by HRTEM and three-step experiment.

Hollow carbon nanospheres have received much interest by virtue of the special shape, low density and large interior void space, allowing their many potential applications. Here we report a template-free route to produce nitrogen-doped hollow porous carbon spheres (denoted as N-HPCSs) through amination reaction of C 60 with alkyl diamines and subsequent carbonization. The size of wall thickness and cavity of hollow carbon spheres could be easily regulated by the chain length of alkyl diamines. After carbonization in Ar atmosphere, the obtained N-HPCSs exhibited excellent activity and recyclability for hydrogenation of aromatic nitro compounds with hydrazine hydrate. The facile approach presented in this work provides a new way for preparation of doped hollow porous carbon spheres.

Autonomous aerial refueling (AAR) is an essential application of unmanned aerial vehicles for both military and civilian domains. In this paper, a hybrid algorithm of the pigeon-inspired optimization (PIO) and lateral inhibition (LI), called LI-PIO, is proposed for image matching problem of AAR. LI is adopted for image pre-processing to enhance the edges and contrast of images. PIO, inspired from the homing characteristics of pigeons, is a novel bio-inspired swarm intelligence algorithm. To demonstrate the effectiveness and feasibility of our proposed algorithm, we make extensive comparative experiments with particle swarm optimization (PSO), particle swarm optimization based on lateral inhibition (LI-PSO), and PIO. It can be concluded from the experimental results that our proposed LI-PIO has excellent performances for image matching problem of AAR, especially in convergent rate and computation speed.

The catalyst Cu3Pt1–Cu2O/SiC was facilely prepared via the in situ reaction of the corresponding compounds supported on SiC in the reaction stream. Cu3Pt1–Cu2O/SiC exhibits excellent catalytic activity for the oxidation of alcohols (conversion of benzyl alcohol and selectivity of benzyl aldehyde are 93% and 98% respectively). The reduction of active Cu2O to inactive Cu0 is the cause behind the deactivation of Cu/SiC. For Cu3Pt1–Cu2O-7/SiC, a Cu2O–Cu3Pt1 alloy formed under the reaction conditions plays an important role in the reaction. Active 5 nm Cu2O nanoparticles are stabilized by the inactive Cu3Pt1 alloy, which was confirmed by control experiments, characterization results and a three-step experiment.

We present an Augmented Reality (AR) visualization and interaction tool for users to control Internet of Things (IoT) devices with hand gestures. Today, smart IoT devices are becoming increasingly ubiquitous with diverse forms and functions, yet most user controls over them are still limited to mobile devices and web interfaces. Recently, AR has been developed rapidly, and provided immersive solutions to enhance user experience of applications in many fields. Its capability to create immersive interactions allows AR to improve the way smart devices are controlled via more direct visual feedback. In this paper, we create a functional prototype of one such system, enabling seamless interactions with sound and lighting systems through the use of augmented hand-controlled interaction panels. To interpret users' intentions, we implement a standard 2D convolution neural network (CNN) for real-time hand gesture recognition and deploy it within our system. Our prototype is also equipped with a simple but effective object detector which can identify target devices within a proper range by analyzing geometric features. We evaluate the performance of our system qualitatively and quantitatively and demonstrate it on two smart devices.