RESUMEN
Formulating expert policies as macro actions promises to alleviate the long-horizon issue via structured exploration and efficient credit assignment. However, traditional option-based multipolicy transfer methods suffer from inefficient exploration of macro action's length and insufficient exploitation of useful long-duration macro actions. In this article, a novel algorithm named enhanced action space (EASpace) is proposed, which formulates macro actions in an alternative form to accelerate the learning process using multiple available suboptimal expert policies. Specifically, EASpace formulates each expert policy into multiple macro actions with different execution times. All the macro actions are then integrated into the primitive action space directly. An intrinsic reward, which is proportional to the execution time of macro actions, is introduced to encourage the exploitation of useful macro actions. The corresponding learning rule that is similar to intraoption Q-learning is employed to improve the data efficiency. Theoretical analysis is presented to show the convergence of the proposed learning rule. The efficiency of EASpace is illustrated by a grid-based game and a multiagent pursuit problem. The proposed algorithm is also implemented in physical systems to validate its effectiveness.
RESUMEN
[-5]One of the greatest challenges for fixed-wing unmanned aircraft vehicles (UAVs) is safe landing. Hereafter, an on-ground deployed visual approach is developed in this paper. This approach is definitely suitable for landing within the global navigation satellite system (GNSS)-denied environments. As for applications, the deployed guidance system makes full use of the ground computing resource and feedbacks the aircraft's real-time localization to its on-board autopilot. Under such circumstances, a separate long baseline stereo architecture is proposed to possess an extendable baseline and wide-angle field of view (FOV) against the traditional fixed baseline schemes. Furthermore, accuracy evaluation of the new type of architecture is conducted by theoretical modeling and computational analysis. Dataset-driven experimental results demonstrate the feasibility and effectiveness of the developed approach.
RESUMEN
This article concentrates on open-source implementation on flying object detection in cluttered scenes. It is of significance for ground stereo-aided autonomous landing of unmanned aerial vehicles. The ground stereo vision guidance system is presented with details on system architecture and workflow. The Chan-Vese detection algorithm is further considered and implemented in the robot operating systems (ROS) environment. A data-driven interactive scheme is developed to collect datasets for parameter tuning and performance evaluating. The flying vehicle outdoor experiments capture the stereo sequential images dataset and record the simultaneous data from pan-and-tilt unit, onboard sensors and differential GPS. Experimental results by using the collected dataset validate the effectiveness of the published ROS-based detection algorithm.
RESUMEN
This paper presents a new algorithm for extrinsically calibrating a multi-sensor system including multiple cameras and a 2D laser scanner. On the basis of the camera pose estimation using AprilTag, we design an AprilTag array as the calibration target and employ a nonlinear optimization to calculate the single-camera extrinsic parameters when multiple tags are in the field of view of the camera. The extrinsic parameters of camera-camera and laser-camera are then calibrated, respectively. A global optimization is finally used to refine all the extrinsic parameters by minimizing a re-projection error. This algorithm is adapted to the extrinsic calibration of multiple cameras even if there is non-overlapping field of view. For algorithm validation, we have built a micro-aerial vehicle platform with multi-sensor system to collect real data, and the experiment results confirmed that the proposed algorithm yields great performance.
