Co-Training has achieved significant success in the field of semi-supervised learning(SSL); however, the homogenization phenomenon, which arises from multiple models tending towards similar decision boundaries, remains inadequately addressed. To tackle this issue, we propose a novel algorithm called β-FFT from the perspectives of data processing and training structure. In data processing, we apply diverse augmentations to input data and feed them into two sub-networks. To balance the training instability caused by different augmentations during consistency learning, we introduce a nonlinear interpolation technique based on the Fast Fourier Transform (FFT). By swapping low-frequency components between variously augmented images, this method not only generates smooth and diverse training samples that bridge different augmentations but also enhances the model’s generalization capability while maintaining consistency learning stability. In training structure, we devise a differentiated training strategy to mitigate homogenization in co-training. Specifically, we use labeled data for additional training of one model within the co-training framework, while for unlabeled data, we employ linear interpolation based on the Beta(β) distribution as a regularization technique in additional training. This approach allows for more efficient utilization of limited labeled data and simultaneously improves the model’s performance on unlabeled data, optimizing overall system performance. Code is available at the following link. https://github.com/Xi-Mu-Yu/beta-FFT.

Learning robotic manipulation skills from vision is a promising approach for developing robotics applications that can generalize broadly to real-world scenarios. As such, many approaches to enable this vision have been explored with fruitful results. Particularly, object-centric representation methods have been shown to provide better inductive biases for skill learning, leading to improved performance and generalization. Nonetheless, we show that object-centric methods can struggle to learn simple manipulation skills in multi-object environments.Thus, we propose DOCIR, an object-centric framework that introduces a disentangled representation for objects of interest, obstacles, and robot embodiment. We show that this approach leads to state-of-the-art performance for learning pick and place skills from visual inputs in multi-object environments and generalizes at test time to changing objects of interest and distractors in the scene. Furthermore, we show its efficacy both in simulation and zero-shot transfer to the real world.

Self-supervised skill learning aims to acquire useful behaviors that leverage the underlying dynamics of the environment. Latent variable models, based on mutual information maximization, have been successful in this task but still struggle in the context of robotic manipulation. As it requires impacting a possibly large set of degrees of freedom composing the environment, mutual information maximization fails alone in producing useful and safe manipulation behaviors. Furthermore, tackling this by augmenting skill discovery rewards with additional rewards through a naive combination might fail to produce desired behaviors. To address this limitation, we introduce SLIM, a multi-critic learning approach for skill discovery with a particular focus on robotic manipulation. Our main insight is that utilizing multiple critics in an actor-critic framework to gracefully combine multiple reward functions leads to a significant improvement in latent-variable skill discovery for robotic manipulation while overcoming possible interference occurring among rewards which hinders convergence to useful skills. Furthermore, in the context of tabletop manipulation, we demonstrate the applicability of our novel skill discovery approach to acquire safe and efficient motor primitives in a hierarchical reinforcement learning fashion and leverage them through planning, significantly surpassing baseline approaches for skill discovery.

The robotic manipulation of deformable objects, such as clothes and fabric, is known as a complex task from both the perception and planning perspectives. Indeed, the stochastic nature of the underlying environment dynamics makes it an interesting research field for statistical learning approaches and neural policies. In this work, we introduce a novel attention-based neural architecture capable of solving a smoothing task for such objects by means of a single robotic arm. To train our network, we leverage an oracle policy, executed in simulation, which uses the topological description of a mesh of points for representing the object to smooth. In a second step, we transfer the resulting behavior in the real world with imitation learning using the cloth point cloud as decision support, which is captured from a single RGBD camera placed egocentrically on the wrist of the arm. This approach allows fast training of the real-world manipulation neural policy while not requiring scene reconstruction at test time, but solely a point cloud acquired from a single RGBD camera. Our resulting policy first predicts the desired point to choose from the given point cloud and then the correct displacement to achieve a smoothed cloth. Experimentally, we first assess our results in a simulation environment by comparing them with an existing heuristic policy, as well as several baseline attention architectures. Then, we validate the performance of our approach in a real-world scenario. Project website: link

Seismic migration produces the migrated images of subsurface media using seismic data, which is important for geophysical exploration. However, the adjoint-based migration methods may produce a blurry image, convolved by a Hessian matrix. To address this problem, we propose a Hessian-assisted iterative self-training learning (HAISTL) method aimed at approximating the inverse Hessian matrix and deblurring the migrated image. First, we train a long short-term (LSTM) network using labeled images and use it as a teacher network to generate pseudolabels for the unlabeled images. Subsequently, we integrate the demigration and migration operators to identify the pseudolabels with high confidence levels and construct a dataset containing both the true and pseudolabels. The dataset is then used to train a student network with the injection of model noise into the network. Finally, we regard the student network as a new teacher and repeat the process in an iterative STL framework. We demonstrate the effectiveness of our proposed method using two synthetic datasets and field data. Compared with the supervised learning (SL) method, the proposed method exhibits superior generalization capabilities. This advantage stems from the incorporation of the demigration and migration operators, providing a valuable prior for the inverse Hessian matrix in training the model. In contrast to the model-driven least-squares migration (LSM) methods, the proposed method yields high-resolution images with significantly reduced computational costs. However, it may be less effective in recovering small-scale structures when confronted with an extremely limited number of labels.

In image processing, the detection of keypoint plays a significant role in foundation work of many image applications. Most widely distributed features in images are corners. Among the famous corner detectors, Harris detector has shown its excellent performance. However, when coming up with SAR image, the speckle noise may severely influence the performance of Harris. And Harris is seldom used in SAR image processing. In this paper, we take advantage of the premium properties of the Bilateral Filter which is robust to speckle noise while preserving the details. Then we propose a new corner detector called bf-Harris. We study the performance of the proposed detector and compare it to several existing approaches. The result shows the algorithm has an excellent performance.