While modern visual recognition systems have made significant advancements, many continue to struggle with the open problem of learning from few exemplars. This paper focuses on the task of object detection in the setting where object classes follow a natural long-tailed distribution. Existing methods for long-tailed detection resort to external ImageNet labels to augment the low-shot training instances. However, such dependency on a large labeled database has limited utility in practical scenarios. We propose a versatile and scalable approach to leverage optional unlabeled images, which are easy to collect without the burden of human annotations. Our SimLTD framework is straightforward and intuitive, and consists of three simple steps: (1) pre-training on abundant head classes; (2) transfer learning on scarce tail classes; and (3) fine-tuning on a sampled set of both head and tail classes. Our approach can be viewed as an improved head-to-tail model transfer paradigm without the added complexities of meta-learning or knowledge distillation, as was required in past research. By harnessing supplementary unlabeled images, without extra image labels, SimLTD establishes new record results on the challenging LVIS v1 benchmark across both supervised and semi-supervised settings.

This paper presents a real-time adaptive trajectory prediction framework for cooperative unmanned aerial vehicle (UAV) swarms engaged in dynamic target tracking within cluttered environments. The proposed system introduces a novel artificial intelligence (AI)-based control architecture combining fuzzy inference, neuro-emotional learning, and distributed multi-agent coordination. At the core of the approach is a Bidirectional Fuzzy Brain Emotional Learning Prediction (BFBEL-P) model, which integrates fuzzy logic and an online adaptive neural structure to enable trajectory forecasting without pre-training or prior knowledge of the environment. From an engineering perspective, this AI model is deployed in UAV swarm navigation, where robust decision-making, predictive coordination, and obstacle avoidance are essential for target interception missions. In contrast to conventional prediction methods, such as curve fitting, nonlinear model predictive control (MPC), and deep learning-based Long Short-Term Memory (LSTM) networks, the BFBEL-P framework offers fast convergence, low computational cost, and high adaptability. The system incorporates multi-threaded data fusion across the swarm to achieve consensus-driven predictions and maintain situational awareness, even under sensor failures or occlusions. Simulation results show that BFBEL-P improves short-term prediction accuracy by 82.2%, reduces prediction time by 15%, and achieves a 100% tracking success rate across benchmark scenarios. These results establish BFBEL-P as a reliable AI technique for distributed control in real-world UAV swarm applications, offering a promising tool for search-and-rescue, surveillance, and defense operations.

Few-shot object detection (FSOD) is a challenging problem aimed at detecting novel concepts from few exemplars. Existing approaches to FSOD all assume abundant base labels to adapt to novel objects. This paper studies the new task of semi-supervised FSOD by considering a realistic scenario in which both base and novel labels are simultaneously scarce. We explore the utility of unlabeled data within our proposed label-efficient detection framework and discover its remarkable ability to boost semi-supervised FSOD by way of region proposals. Motivated by this finding, we introduce SoftER Teacher, a robust detector combining pseudo-labeling with consistency learning on region proposals, to harness unlabeled data for improved FSOD without relying on abundant labels. Rigorous experiments show that SoftER Teacher surpasses the novel performance of a strong supervised detector using only 10% of required base labels, without catastrophic forgetting observed in prior approaches. Our work also sheds light on a potential relationship between semi-supervised and few-shot detection suggesting that a stronger semi-supervised detector leads to a more effective few-shot detector.

Ensuring robust and precise tracking control in the presence of uncertain multi-input-multi-output (MIMO) system dynamics and environmental variations is a significant challenge in the field of robust and adaptive control theory. While fuzzy control strategies have demonstrated good tracking performance in normal conditions, designing and tuning fuzzy controllers can be a challenging task in highly uncertain environments. In this study, we investigate a novel approach that combines robust nonlinear negative-imaginary (NI) systems theory with a self-adaptive fuzzy control scheme and the Lyapunov synthesis to develop a robust adaptive negative-imaginary-fuzzy (RANIF) control scheme. We optimize the critical parameters of the proposed fuzzy system using a self-tuning technique with a proportional-derivative sliding manifold. Furthermore, unlike the existing adaptive fuzzy control methods, we propose a small number of membership functions and systematically derive the fuzzy rules by employing Lyapunov, nonlinear NI, and dissipativity theories, which simplify the tuning process, work out the matter of "explosion of complexity," and reduce computational complexity. We demonstrate the global stability of the closed-loop system using nonlinear NI theory. To evaluate the effectiveness of our proposed approach, we present simulation results for two examples involving uncertain MIMO second-order Euler-Lagrange systems. These systems, known for their capacity to represent a diverse range of practical physical systems, serve as suitable testbeds for our methodology. Our results show that RANIF outperforms other control methods, such as nonlinear strictly NI-Fuzzy, fuzzy-logic control, model predictive control, and conventional PID control, in terms of robustness to disturbances and inestimable faults, trajectory tracking performance, and computational complexity.

Quadrotors are one of the popular unmanned aerial vehicles (UAVs) due to their versatility and simple design. However, the tuning of gains for quadrotor flight controllers can be laborious, and accurately stable control of trajectories can be difficult to maintain under exogenous disturbances and uncertain system parameters. This article introduces a novel robust adaptive control synthesis methodology for a quadrotor robot's attitude and altitude stabilization. The proposed method is based on the fuzzy reinforcement learning and strictly negative imaginary (SNI) property. The first stage of our control approach is to transform a nonlinear quadrotor system into an equivalent negative-imaginary (NI) linear model by means of the feedback linearization (FL) technique. The second phase is to design a control scheme that adapts online the SNI controller gains via fuzzy Q -learning. The performance of the designed controller is compared with that of a fixed-gain SNI controller, a fuzzy-SNI controller, and a conventional PID controller in a series of numerical simulations. Furthermore, the proofs for the stability of the proposed controller and the adaptive laws are provided using the NI theorem.

This paper proposes an innovative approach to coverage path planning and obstacle avoidance for multiple Unmanned Ground Vehicles (UGVs) in a changing environment, taking into account constraints on the time, path length, number of UGVs and obstacles. Our approach leverages deformable virtual leader-follower formations to enable UGVs to adapt their formation based on both planned and real-time sensor data. A hierarchical block algorithm is employed to identify areas in the environment where UGV formations can spread out to meet time and budget constraints. Additionally, we introduce a novel control scheme that allows each UGV to generate a local steering force to dodge any static and mobile obstacles based on the closest safe angle. Results from simulations and real UGV experiments demonstrate that our approach achieves a higher coverage percentage than rule-based and reactive swarming approaches without planning. Our approach offers a promising solution for efficient coverage path planning and obstacle avoidance in complex environments with multiple UGVs.

Recent years have seen flourishing research on both semi-supervised learning and 3D room layout reconstruction. In this work, we explore the intersection of these two fields to advance the research objective of enabling more accurate 3D indoor scene modeling with less labeled data. We propose the first approach to learn representations of room corners and boundaries by using a combination of labeled and unlabeled data for improved layout estimation in a 360° panoramic scene. Through extensive comparative experiments, we demonstrate that our approach can advance layout estimation of complex indoor scenes using as few as 20 labeled examples. When coupled with a layout predictor pre-trained on synthetic data, our semi-supervised method matches the fully supervised counterpart using only 12% of the labels. Our work takes an important first step towards robust semi-supervised layout estimation that can enable many applications in 3D perception with limited labeled data.

Formation control techniques have been widely implemented in networked multirobot systems. In this article, we present a novel framework for swarm multiagent systems based on the relative-position output feedback consensus supported with the new concept of adaptive strictly negative imaginary consensus controllers, leveraging the learning capability of artificial neural networks. For experimental validation, we consider the case of two quadcopters moving together while carrying a dynamic load. We employ Kharitonov's theorem to study the stability of the proposed adaptive control systems. Finally, a rigorous real-time experimental study is conducted to highlight the merits of the proposed formation control algorithms.

We examine two fundamental tasks associated with graph representation learning: link prediction and semi-supervised node classification. We present a novel autoencoder architecture capable of learning a joint representation of both local graph structure and available node features for the multi-task learning of link prediction and node classification. Our autoencoder architecture is efficiently trained end-to-end in a single learning stage to simultaneously perform link prediction and node classification, whereas previous related methods require multiple training steps that are difficult to optimize. We provide a comprehensive empirical evaluation of our models on nine benchmark graph-structured datasets and demonstrate significant improvement over related methods for graph representation learning. Reference code and data are available at https://github.com/vuptran/graph-representation-learning.