Abstract-This work focuses on optimal routing for two camera-equipped UAVs cooperatively tracking a single target moving on the ground. The UAVs are small fixed-wing aircraft cruising at a constant speed and fixed altitude; consequently, the vehicles are modeled as planar Dubins vehicles. A perspective transformation, relating the image-plane measurements to the ground, allows derivation of the geolocation (target localization) error covariance. Using dynamic programming, we compute optimal coordinated control policies which minimize the fused geolocation error covariance. A surprising result, and the main contribution of this work, is that the dominant factor governing the optimal UAV routes is coordination of the distances to the target, not of the viewing directions as is traditionally assumed.
Abstract-This work focuses on enabling multiple UAVs to flock together in order to distribute and collectively perform a given sensing task. Flocking is performed in a leader-follower fashion, and the leader is assumed to already have an effective control policy for the particular task. The UAVs are small fixedwing aircraft cruising at a constant speed and fixed altitude, but experience stochasticity in their dynamics. Accordingly, the control problem for each follower is addressed in the context of stochastic optimal control, wherein the cost is a function of distance and heading with respect to the leader. The problem is solved offline via dynamic programming to minimize the expected cost over a finite horizon and generate a receding horizon optimal control policy. This flocking algorithm was successfully applied in the field, where three camera-equipped UAVs flocked together to perform vision-based target tracking. The experimental results verify the efficacy of the approach and show the benefits of flocking with multiple UAVs to distribute sensing tasks, which include a dramatic reduction in overall sensing error and robustness to individual sensor faults.
This paper considers the problem of a small, fixed-wing UAV equipped with a gimbaled camera autonomously tracking an unpredictable moving ground vehicle. Thus, the UAV must maintain close proximity to the ground target and simultaneously keep the target in its camera's visibility region. To achieve this objective robustly, two novel optimizationbased control strategies are developed. The first assumes an evasive target motion while the second assumes a stochastic target motion. The resulting optimal control policies have been successfully flight tested, thereby demonstrating the efficacy of both approaches in a real-world implementation and highlighting the advantages of one approach over the other.
In this article, we present a monocular visual reactive navigation system capable of navigating at high speeds, without GPS, in unknown complex cluttered environments. The system, called R-ADVANCE (Rapid Adaptive Prediction for Vision-based Autonomous Navigation, Control, and Evasion), consists of a set of biologically inspired visual perception and reactive control algorithms that provide low-computation reactive obstacle avoidance while navigating at high speeds in search of a goal object. These algorithms, along with basic planning, and augmented with lowprecision visual odometry, were implemented on a micro unmanned aerial vehicle and tested in a number of challenging environments. While each of the individual algorithmic and hardware elements has been previously studied in limited environments, this work is the first time that these novel components have been integrated and flight-tested. To achieve fast flight, an NVIDIA Tegra TK1 was used as the main processor, allowing us to parallelize the system to process 1280 × 720 video streams at 40 fps, reaching flight speeds up to 19 m/s (≈68 km/h) or 42 mph.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.