A compliant control model based on reinforcement learning (RL) is proposed to allow robots to interact with the environment more effectively and autonomously execute force control tasks. The admittance model learns an optimal adjustment policy for interactions with the external environment using RL algorithms. The model combines energy consumption and trajectory tracking of the agent state using a cost function. Therein, an Unmanned Aerial Vehicle (UAV) can operate stably in unknown environments where interaction forces exist. Furthermore, the model ensures that the interaction process is safe, comfortable, and flexible while protecting the external structures of the UAV from damage. To evaluate the model performance, we verified the approach in a simulation environment using a UAV in three external force scenes. We also tested the model across different UAV platforms and various low-level control parameters, and the proposed approach provided the best results.

Binocular depth estimation is a fundamental problem in computer vision. Learning-based models have achieved significant performance improvements on public datasets in recent years. Our study finds that the performance of the current state-of-the-art deep learning-based models deteriorates significantly in distant areas. We point out that these deep learning-based models suffer from a scale shrinkage problem. Specifically, the predicted depth value ratio to the ground truth decreases as depth increases. Such a phenomenon is not conducive to the path planning and navigation of intelligent agents in outdoor scenes. We analyze the reasons for the scale shrinkage problem and give a simple and effective method. Our method employs a two-stage fine-tuning strategy and appropriately fuses the predictions of the two-stage models. The method does not reduce the prediction accuracy in close areas and significantly improves the accuracy of the models in distant areas. On the KITTI stereo 2015 dataset, our method can reduce the absolute relative difference by about 6% and the root-mean-square error (RMSE) by about 10%.

Satellite image dehazing aims at precisely retrieving the real situations of the obscured parts from the hazy remote sensing (RS) images, which is a challenging task since the hazy regions contain both ground features and haze components. Many approaches of removing haze focus on processing multi-spectral or RGB images, whereas few of them utilize multi-sensor data. The multi-sensor data fusion is significant to provide auxiliary information since RGB images are sensitive to atmospheric conditions. In this paper, a dataset called SateHaze1k is established and composed of 1200 pairs clear Synthetic Aperture Radar (SAR), hazy RGB, and corresponding ground truth images, which are divided into three degrees of the haze, i.e. thin, moderate, and thick fog. Moreover, we propose a novel fusion dehazing method to directly restore the haze-free RS images by using an end-to-end conditional generative adversarial network(cGAN). The proposed network combines the information of both RGB and SAR images to eliminate the image blurring. Besides, the dilated residual blocks of the generator can also sufficiently improve the dehazing effects. Our experiments demonstrate that the proposed method, which fuses the information of different sensors applied to the cloudy conditions, can achieve more precise results than other baseline models.

Affine-invariant feature matching plays an important role in many robot vision applications, such as robot visual navigation, object detection, visual tracking and visual SLAM, etc. In the early stages, invariant keypoints are used to detect the affine transformation. But the accuracy is very low. In recent years, some people introduce SIFT method into robot vision field, which greatly enhances the accuracy. But it is too time-consuming to meet the requirements of real-time robot vision applications. In this paper, we propose a novel learning-based feature matching approach to address the problem. First, it uses a fast algorithm to extract keypoints. Then, our method identifies keypoints that belong to different objects or background by color and texture representation. The keypoints are clustered into corresponding groups. At last, a two-stage multilayer ferns classifier is trained to recognize the local patches and get the estimate of viewpoint. We test our approach on public datasets and apply it in a visual SLAM application. The result demonstrates that our method can provide robust and powerful matching ability. Even on some difficult matching cases, it also performs remarkably well. Further more, because there is no need to compute descriptors for the image, our method is very fast at run-time.

While local patches recognition is a key component of modern approaches to affine transformation detection and object detection, existing learning-based approaches just identify the patches based on a set of randomly picked and combined binary features, which will lose some strong correlations between features and can not provide stable and remarkable identification ability. In this paper, we proposed a method that select and organize the features in a Multilayer Ferns structure, and show that it is both faster in the run-time processing and more powerful in the identification ability than state-of-the-art ad hoc approaches.

The stock market is considered as a high complex and dynamic system. Many machine learning and data mining technologies are used for stock analysis, but it still leaves an open question about how to integrate these methods with the plentiful knowledge and techniques accumulated in stock investment which are critical to the successful stock analysis. In this paper, we propose an intelligent stock trading system by combining support vector machine (SVM) algorithm and box theory of stock. The box theory believes a successful stock buying/selling generally occurs when the price effectivley breaks out the original oscillation box into another new box. In the system, support vector machine algorithm is utilized to make forecasts of the top and bottom of the oscillation box. Then a trading strategy based on the box theory is constructed to make trading decisions. The different stock movement patterns, i.e. bull, bear and fluctuant market, are used to test the feasibility of the system. The experiments on S&P500 components show a promising performance is achieved.

In this paper, neural network based ensemble learning methods are introduced in predicting activities of COX-2 inhibitors in Chinese medicine Quantitative Structure-Activity Relationship (QSAR) research. Three different ensemble learning methods: bagging, boosting and random subspace are tested using neural networks as basic regression rules. Experiments show that all three methods, especially boosting, are fast and effective ways in the activity prediction of Chinese medicine QSAR research, which is generally based on a small amount of training samples.