Search for: All records

Distributed multi-agent unmanned aerial systems (UAS) have the potential to be heavily utilized in environmental monitoring, especially in wetland monitoring. Deep active learning algorithms provide key tools to analyze the sensed images captured during monitoring and interpret them precisely. However, these algorithms demand significant computational resources that limit their use with distributed UAS. In this paper, we propose a novel algorithm for consensus-enabled active learning that drastically reduces the computational demand while increasing the overall model accuracy. Once each of the UAS obtains a labeled subset of images through active learning, we update the weights of the model for three epochs only on the new images to reduce the computational cost, allowing for an increased operational time. The group of UAS communicates the model weights instead of the raw data and then leverages consensus to agree on updated weights. The consensus step mitigates the impact on weights caused by the updates and generalizes the knowledge of each individual UAS to the whole system, which results in increased model accuracy. Our method achieved an average of 11.15% increase in accuracy over 25 acquisition iterations whilst utilizing only 16.8% of the processor time compared to the centralized method of active learning. 

The primary objective of this study was to utilize state-of-the-art deep learning-based monocular depth estimation models to assist UAS pilots in rainforest canopy data collection and navigation. Monocular depth estimation models provide a complementary technique to other depth measurement and estimation techniques to extend the range and improve mea- surements. Several state-of-the-art models were evaluated using a novel dataset composed of data from a simulated rainforest environment. In the evaluation, MiDaS outperformed the other models, and a segmentation pipeline was designed using this model to identify the highest areas of the canopies. The segmen- tation pipeline was evaluated using 1080p and 360p input videos from the simulated rainforest dataset. It was able to achieve an IoU of 0.848 and 0.826 and an F1 score of 0.915 and 0.902 at each resolution, respectively. We incorporated the proposed depth-estimation-based segmentation pipeline into an example application and deployed it on an edge system. Experimental results display the capabilities of a UAS using the segmentation pipeline for rainforest data collection. 

Unmanned aerial vehicles (UAVs) are becoming more common, presenting the need for effective human-robot communication strategies that address the unique nature of unmanned aerial flight. Visual communication via drone flight paths, also called gestures, may prove to be an ideal method. However, the effectiveness of visual communication techniques is dependent on several factors including an observer's position relative to a UAV. Previous work has studied the maximum line-of-sight at which observers can identify a small UAV [1]. However, this work did not consider how changes in distance may affect an observer's ability to perceive the shape of a UAV's motion. In this study, we conduct a series of online surveys to evaluate how changes in line-of-sight distance and gesture size affect observers' ability to identify and distinguish between UAV gestures. We first examine observers' ability to accurately identify gestures when adjusting a gesture's size relative to the size of a UAV. We then measure how observers' ability to identify gestures changes with respect to varying line-of-sight distances. Lastly, we consider how altering the size of a UAV gesture may improve an observer's ability to identify drone gestures from varying distances. Our results show that increasing the gesture size across varying UAV to gesture ratios did not have a significant effect on participant response accuracy. We found that between 17 m and 75 m from the observer, their ability to accurately identify a drone gesture was inversely proportional to the distance between the observer and the drone. Finally, we found that maintaining a gesture's apparent size improves participant response accuracy over changing line-of-sight distances. 

This article presents an understanding of naive users’ perception of the communicative nature of unmanned aerial vehicle (UAV) motions refined through an iterative series of studies. This includes both what people believe the UAV is trying to communicate, and how they expect to respond through physical action or emotional response. Previous work in this area prioritized gestures from participants to the vehicle or augmenting the vehicle with additional communication modalities, rather than communicating without clear definitions of the states attempting to be conveyed. In an attempt to elicit more concrete states and better understand specific motion perception, this work includes multiple iterations of state creation, flight path refinement, and label assignment. The lessons learned in this work will be applicable broadly to those interested in defining flight paths, and within the human-robot interaction community as a whole, as it provides a base for those seeking to communicate using non-anthropomorphic robots. We found that the Negative Attitudes towards Robots Scale (NARS) can be an indicator of how a person is likely to react to a UAV, the emotional content they are likely to perceive from a message being conveyed, and it is an indicator for the personality characteristics they are likely to project upon the UAV. We also see that people commonly associate motions from other non-verbal communication situations onto UAVs. Flight specific recommendations are to use a dynamic retreating motion from a person to encourage following, use a perpendicular motion to their field of view for blocking, simple descending motion for landing, and to use either no motion or large altitude changes to encourage watching. Overall, this research explores the communication from the UAV to the bystander through its motion, to see how people respond physically and emotionally.