Threading the Needle: Test and Evaluation of Early Stage UAS Capabilities to Autonomously Navigate GPS-Denied Environments in the DARPA Fast Lightweight Autonomy (FLA) Program

TL;DR

The paper documents DARPA's Fast Lightweight Autonomy Phase 1 testing of high-speed, GPS-denied UAS navigation on a common hardware platform, focusing on autonomy and sensing rather than hardware. It introduces a rigorous testing methodology using AprilTags fiducials, mission-files with relative bearings, a camera-convergence success criterion within a cylindrical target region, and concrete metrics for goal attainment and return, plus a human teleoperation baseline. Across four escalating experiments, the results show meaningful progress toward autonomous flight at up to $20~\mathrm{m/s}$, with best-case goal attainments of up to $36\%$ and mixed success in returning to start, highlighting both the potential and the challenges of GPS-denied autonomous navigation. The work provides a critical benchmarking framework that influenced subsequent advances in onboard autonomy, visual-inertial odometry, and monocular navigation in cluttered, GPS-denied environments.

Abstract

The DARPA Fast Lightweight Autonomy (FLA) program (2015 - 2018) served as a significant milestone in the development of UAS, particularly for autonomous navigation through unknown GPS-denied environments. Three performing teams developed UAS using a common hardware platform, focusing their contributions on autonomy algorithms and sensing. Several experiments were conducted that spanned indoor and outdoor environments, increasing in complexity over time. This paper reviews the testing methodology developed in order to benchmark and compare the performance of each team, each of the FLA Phase 1 experiments that were conducted, and a summary of the Phase 1 results.

Figures (7)

• Figure 1: The three performing teams of the DARPA FLA program.
• Figure 2: Testing artifacts at the start position and goal.
• Figure 3: Example contents of a mission file: (clockwise from left) overhead map with trajectory, AprilTags, and target object images.
• Figure 4: Diagrams of the camera convergence method.
• Figure 5: Experiments conducted during Phase 1 of the DARPA FLA program.