Aerial Object Trajectory Classification

Onboard collision avoidance is needed to enable safe, autonomous flight operations for NASA projects such as Advanced Air Mobility (AAM), as well as many commercial applications. Real-time aerial object classification will improve onboard collision avoidance algorithm decision making and may reduce unnecessary activation of avoidance systems. This work trains an aircraft trajectory classifier using trajectories from flight controller logs and tests the classifier using RADAR collected trajectories during air-to-air experiments and ground-to-air experiments. In contrast to RADAR data, these flight controller logs are relatively abundant, which makes the possibility of substituting flight data for RADAR data an attractive, cost-effective option. The Support Vector Machine (SVM) model developed in this work achieved a 79.7% classification accuracy on the first second of radar trajectories of general aviation (GA) aircraft, multirotor small unmanned aircraft systems (sUAS), and fixed wing sUAS. Findings from this effort show that it is feasible to classify sensor collected trajectories using a classifier trained flight controller data.

Associated Publication:
H Holbrook, P Garrett, N Behari, C Dolph, C I. Morris, and G Szatkowski. "Aerial Object Trajectory Classification by Training on Flight Controller Data and Testing on RADAR Generated Tracks." In AIAA SCITECH 2022 Forum, p. 0496. 2022.

Data Overview

1. Radar Data: The radar received detections at a frequency of 9 Hz and used a Kalman filter in order to assign detections to tracks. Additionally, to avoid false positives, the radar would only return a track once it hit a certain threshold of confidence that the track was indeed an object. It was in this track form that the data was returned [1]. A given track provides the following information about the corresponding object. This information was updated approximately 9 times per second: 1) X, Y, and Z Estimated Position (in meters) 2) X, Y, and Z Estimated Velocity (in m/s) 3) Estimated radar Cross-Section (RCS).

2. Flight Controller Data: The flight controller logs had a variety of different data sources that provided position and velocity measurements that were specific to the aircraft. The data source chosen was an extended Kalman filter that not only takes the GPS as input, but also takes inputs from the other available sensors on the aircraft into account, potentially including inertial measurement units, gyroscopes, accelerometers, and barometers. This extended Kalman filter feature was selected as the data source for this project due to its increased number of sensors being utilized, leading to more robust measurements. When this feature was not available, the raw GPS log was used instead. The information provided by the Kalman filter or the raw GPS was as follows: 1) Latitude and Longitude (in degrees) 2) Velocity North, East, and Down (in m/s) 3) Altitude (in meters).

[1] Dolph, C. V., Szatkowski, G. N., Holbrook, H., Morris, C. M., Ticatch, L. A., Malekpour, M. R., and McSwain, R. G., “Aircraft Classification Using RADAR from small Unmanned Aerial Systems for Scalable Traffic Management Emergency Response Operations,” 2021. https://doi.org/10.2514/6.2021-2331.

This data is valuable for aircraft trajectory classification. This data benefits onboard collision avoidance and generally supports autonomous flight operations.

This is early/mid-stage (TRL 4) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.