Precision Enhancement of Airdrop Releases through Learning (PEARL)

Period of Performance: 07/13/2015 - 03/30/2016


Phase 1 SBIR

Recipient Firm

Aurora Flight Sciences Corp.
9950 Wakeman Drive Array
Manassas, VA 20110
Principal Investigator


ABSTRACT:Airdrop accuracy is of paramount importance since airdrop systems landing in unintended locations could create internationally damaging outcomes. There is an opportunity to compensate for local effects by learning their impact on the airdrop trajectory from previous missions. Aurora Flight Sciences is teaming with Boston University to apply deep learning techniques to historical airdrop data at particular drop zones to determine site- and mission-specific biases and increase accuracy at that drop zone. Deep learning can identify complex multi-layer relationships within data and discover patterns without any prior knowledge, enabling detection of trends that are only present when multiple conditions are present. Although deep learning calculations are computationally expensive, they can be run offline to generate a database that can be accessed online based on real-time mission information. The algorithms can ingest additional data as it is collected to update and refine outputs for even greater accuracy. The conceptual tool developed in this Phase I SBIR will plan flight paths and computed air release points based on knowledge learned from past airdrops with the ultimate goal of showing the feasibility of using historical data to improve accuracy at a particular drop zone.BENEFIT:Recent operations in Afghanistan and Syria have increased the interest in precision airdrop as a low cost and safe method of tactical resupply, and mission commanders are continuing to seek ways to improve airdrop accuracy and reliability. Calculation of site- and mission-specific airdrop adjustments will break through the current barrier in airdrop accuracy created by limitations in weather forecasting. Increases in airdrop accuracy reduce risk to our ground and air forces, increase bundle recovery rates, and enable missions at more highly constrained drop zones. Further, this deep learning based technique can be extended to other airdrop system types including humanitarian, dispersion, and guided systems. The tool developed in this effort can be integrated into multiple airdrop mission planning programs including the Consolidated Airdrop Tool, XDrop, and PARANAVSYS. Additionally, it can be extended to high altitude sonobuoy deployment for the Navy.