1,721,160 research outputs found
Obstacle avoidance for UAVs via imitation learning from human data with 3D-convolutional neural networks
No full textVision-based obstacle avoidance for UAVs via imitation learning with sequential neural networks
No full textThis paper explores the feasibility of a framework for vision-based obstacle avoidance techniques that can be applied to unmanned aerial vehicles, where such decision-making policies are trained upon supervision of actual human flight data. The neural networks are trained based on aggregated flight data from human experts, learning the implicit policy for visual obstacle avoidance by extracting the necessary features within the image. The images and flight data are collected from a simulated environment provided by Gazebo, and Robot Operating System is used to provide the communication nodes for the framework. The framework is tested and validated in various environments with respect to four types of neural network including fully connected neural networks, two- and three-dimensional convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Among the networks, sequential neural networks (i.e., 3D-CNNs and RNNs) provide the better performance due to its ability to explicitly consider the dynamic nature of the obstacle avoidance problem
Persistent standoff tracking guidance using constrained particle filter for multiple UAVs
No full textThis paper presents a new standoff tracking framework of a moving ground target using UAVs with limited sensing capabilities and motion constraints. To maintain persistent track of the target even in case of target loss for a certain period, this study predicts the target existence area using the particle filter and produces control commands that ensure that all predicted particles can stay within the field-of-view of the UAV sensor at all times. To improve target position prediction and estimation accuracy, the road information is incorporated into the constrained particle filter where the road boundaries are modelled as inequality constraints. Both Lyapunov vector field guidance and nonlinear model predictive control-based methods are applied, and the characteristics of them are compared using numerical simulations
Heterogeneous mission planning for a single unmanned aerial vehicle (UAV) with attention-based deep reinforcement learning
Full text linkLarge-scale and complex mission environments require unmanned aerial vehicles (UAVs) to deal with various types of missions while considering their operational and dynamic constraints. This article proposes a deep learning-based heterogeneous mission planning algorithm for a single UAV. We first formulate a heterogeneous mission planning problem as a vehicle routing problem (VRP). Then, we solve this by using an attention-based deep reinforcement learning approach. Attention-based neural networks are utilized as they have powerful computational efficiency in processing the sequence data for the VRP. For the input to the attention-based neural networks, the unified feature representation on heterogeneous missions is introduced, which encodes different types of missions into the same-sized vectors. In addition, a masking strategy is introduced to be able to consider the resource constraint (e.g., flight time) of the UAV. Simulation results show that the proposed approach has significantly faster computation time than that of other baseline algorithms while maintaining a relatively good performance
Prescribed performance adaptive finite-time control for uncertain horizontal platform systems
No full textThis paper presents a new approach to the design of prescribed performance adaptive control for uncertain horizontal platform systems with the finite-time convergence. Following an appropriate performance function and error transformation, a new adaptive control law is proposed by using a novel integral non-singular terminal sliding mode surface. The proposed approach simultaneously guarantees that (i) the transient responses of the closed-loop system possess some advanced properties such as the existence of the prespecified lower bound of the convergence rate and of the pre-established upper bound of the maximum overshoot; and (ii) the finite-time convergence of the state trajectories/tracking errors to zero. The global stability and finite-time convergence are strictly analyzed. The proposed method is clarified and verified through two numerical simulation examples. (C) 2020 ISA. Published by Elsevier Ltd. All rights reserved
Target Classification and Prediction of Unguided Rocket Trajectories Using Deep Neural Networks
No full textReceding-horizon RRT-Infotaxis for autonomous source search in urban environments
Full text linkIn emergency situations such as hazardous gas leak, search and estimation for identifying source information, known as source term estimation (STE), in a timely and accurate manner is of significant importance. In real world situations, obstacles such as buildings or barriers not only block the path for search but also interfere the flow of the gas source. For autonomous source search and estimation using a mobile sensor in such obstacle-rich environments, this paper proposes an information-theoretic STE approach by combining a widely-used Infotaxis with the rapidly-exploring random trees (RRT). In particular, the proposed strategy utilizes the receding-horizon RRT concept with a newly designed utility function for determining the next maneuver of a mobile agent to get the best information of the source while avoiding obstacles in urban environments. Numerical simulations in various environments show the superior performance of the proposed approach compared with the original Infotaxis method
Trail navigation with obstacle avoidance using convolutional neural networks
No full textIn this paper, we propose a method of following bike trails while avoiding obstacles using convolutional neural networks (CNN) for an unmanned aerial vehicle (UAV). The direction of the UAV is controlled to follow the given trail, while keeping its position near the center of the trail using the CNN. In addition, to return to the UAV’s original path whenever it goes out of the path due to disturbance, the yaw rates from the CNN are stored and utilized the recent past. To avoid obstacles during the trail navigation, the optical flow estimated from another CNN is used. By integrating these methods, the UAV deals with various situations encountered while traveling on the road, and the proposed approach is verified through simulations using ROS and Gazebo
- …
