The objective of this research is to help Windracers lead the next generation of air transportation through the creation and realisation of autonomous drones. Windracers delivers essential logistics and monitoring to the people and places that need them on time, every time, anywhere. (https://windracers.com/)
Windracers was conceived to provide essential logistical support to humanitarian agencies looking to sustain communities through emergencies where transportation links are poor. Our solution was designed to provide a robust, cost-effective alternative to expensive, crewed aviation for reaching local and remote villages quickly and effectively.
Our essential logistics platform and solution has emerged at the forefront of uncrewed, autonomous aviation and its application and utility is relevant to a wide range of sectors and situations.
Below are the main problems to solve for achieving the m:N operations in the future and ongoing topics are labeled on the right.
We present COVEE, a dataset for Cognitive mOdeling with Video, Electroencephalography, and Eye tracker. COVEE aims to be a valuable resource for advancing cognitive mod- eling research and the real-time estimation of cognitive load (CL) and situation awareness (SA) in complex operational en- vironments. COVEE is collected from 24 subjects conducting two different tasks designed based on the remote operation of unmanned aerial vehicles. In total, more than 50 hours of data are collected, along with self-reported ground truth for CL and SA. To establish benchmarks, extensive experiments and ablation studies are conducted using various modern deep learn- ing architectures and pre-trained large vision models. Insights gained from the experiments and data show the potential of COVEE to be a critical dataset for future cognitive modeling research. Both raw and processed experiment data, along with all codes, are published for the highest reproducibility.
Task 1: The first task is a visual tracking task. In each trial, the subject needs to watch a 10-second video where a red arrow and one or more white arrows move in a constant speed, as shown in the left figure. The subjects are informed that the red arrow represents their own UAV, while the white arrows represent intruders. After each video concludes, all arrows disappear and the subject must use the mouse to click on the monitor to indicate all arrows' latest position and heading. The position of an arrow is marked by the first click, and its heading is determined by the subsequent click, indicating the direction in which the arrow was moving. Each click is marked by a blue box containing a number, which indicates the sequence of the clicks. Each pair of clicks is connected by a blue line, enabling subjects to better visualize their indicated heading. An undo button is provided at the top left corner where the subjects can use to undo their clicks if necessary. All things mentioned above are also included in an instruction video, which is played at the beginning of Task 1. We ask the subjects questions after the instruction video to make sure they fully understand what they need to do in this task and how to indicate the position and heading. After the subject finishes making all clicks, they can click the Next button on the bottom and complete the NASA-TLX survey. Completing the NASA-TLX survey represents the end of one trial in Task 1. To control the task complexity, we control the number of intruders in one trial to be 1, 3, or 5. To minimize the learning effect, the order of the videos is random for each subject. In total, there are 30 trials in Task 1. Additionally, a secondary task is added in the last 15 trials to further increase the task complexity. Details about the design of the secondary task are included in the Secondary Task section.
Task 2: The second task is designed based on the real-world operations utilizing Windracers' cloud control system. In this task, subjects are required to use the cloud control system to: (1) plan a UAV mission from one airport to another by setting waypoints, (2) monitor the UAV during the mission, and (3) set new waypoints to avoid any intruders that may appear during the mission. Each trial lasts between 15 to 20 minutes, depending on the subjects. Prior to the first trial, the subject watches a series of training videos that provide instructions on using the cloud control system. To standardize the subjects' behaviors and facilitate performance assessment, specific guidelines are provided. For instance, the take-off and landing angles must be set between 3 to 5 degrees, and the airspeed must be adjusted to 70 knots during cruising and 45 knots prior to landing. These guidelines are detailed in the training videos and are also available as a hard copy attached to the bottom of the monitor, allowing subjects to reference them as needed. Following the training videos, the subjects are guided through a test session in which they plan and conduct a mission that navigates one UAV around an airport. Trials begin only after we are convinced that the subject has sufficiently mastered the use of the system. To collect the subjects' subjective CL without disrupting the trials, we employ the instantaneous self-assessment (ISA). Once the subject is ready to start the trial, the subject can click the start button. Every minute thereafter, five red boxes, each labeled with a number, appear on the screen, as illustrated in the figure above. The subject needs to click the box that best represents their perceived cognitive load level at that moment. At the end of every trial, the subject needs to complete a NASA-TLX survey (same as the one in Task 1), followed by a 5-minute break. The ISA results serve as the reference label of the subject's CL during the mission, while NASA-TLX survey results are used to assess the subject's overall experience.
Recent advancements in unmanned aerial vehicles (UAVs) bring innovation across different industries and inspire a wide range of applications. Among them, mid-mile delivery stands out for its huge potential in leveraging UAVs to greatly increase efficiency and explore regions that are hard to reach using existing transportation systems. However, due to the properties of UAVs, that is, relatively small size and mass, limited visual and sensor information, and low flight levels, they are generally: 1) susceptible to weather, 2) risky to properties on the ground, and 3) restricted to various airspaces. This paper aims to find the optimal 3D path for UAV mid-mile delivery by considering these factors. Weather forecasts, ground risks, and airspace information are integrated to build costs and constraints for the 3D path planning algorithm. The simulation results show that the generated 3D path can effectively reduce total mission time, fuel consumption, and risk while ensuring restricted airspace and areas are avoided.
In this paper, we aim to find the optimal 3D path for the UAV mid-mile delivery under costs and constraints that are constructed from the weather forecasts, ground risks, and airspace information. The weather forecasts are obtained from the national weather service which provides high resolution gridded weather forecast data at different flight levels. The ground risks are computed based on the population density. The airspace information can obtained from publicly accessible places such as SkyVector. To the best of our knowledge, there are few works in the UAV 3D path planning literature that formulates the problem with all the aforementioned information for UAV mid-mile delivery.
Publication:
C. Deng, W. Sribunma, S. Brunswicker, and I. Hwang, “3D Path Planning With Weather Forecasts, Ground Risks, and Airspace Information for UAV Mid-Mile Delivery,” in Proceedings of the AIAA SCITECH 2025 Forum. AIAA, 2025. (under review)
Publications
C. Deng, W. Sribunma, S. Brunswicker, and I. Hwang, “3D Path Planning With Weather Forecasts, Ground Risks, and Airspace Information for UAV Mid-Mile Delivery,” in Proceedings of the AIAA SCITECH 2025 Forum. AIAA, 2025. (under review)
People
Chuhao Deng, Ph.D. student
Windracers
This material is based upon work supported by Windracers