A drone built by Purdue University researchers followed a runner through an unmapped forest without GPS, weaving between trees and using its onboard computer to predict where the person would reappear. The 44-second demonstration shows a skill that could help rescuers keep track of someone through dense vegetation. Finding a missing person in the first place remains a separate challenge.

Aniket Bera, an Indian-origin computer science professor who directs Purdue’s IDEAS Lab, shared the footage on LinkedIn on October 24. The team designed the aircraft’s carbon-fibre frame and developed the software that tracks the runner and plans the drone’s route. The system makes flight decisions onboard without relying on an internet connection or a map loaded before takeoff.

How the drone tracks a runner behind trees

The drone combines images from a depth-sensing camera with three-dimensional measurements from LiDAR, which uses laser pulses to measure distances. It builds a map of nearby obstacles as it flies. When a trunk blocks the camera’s view, the tracking software uses earlier observations to estimate where the runner is heading and where they may come back into sight. The aircraft can also brake to maintain separation, Bera told TechSpot.

In the LinkedIn post, Bera wrote that “we cannot afford offboard latency” when the aircraft banks to avoid a branch. He said the onboard processor must turn camera data into motor commands within milliseconds while working within the aircraft’s limits on weight, power and heat.

The video’s flight charts prompted questions about the software behind the demonstration. Bera told TechSpot that a label for SUPER, a University of Hong Kong navigation system, came from an earlier comparison script. Purdue uses its own flight planner, alongside the open-source FAST-LIO software to estimate the drone’s position and a modified version of ROG-Map to map obstacles. Bera said a paper describing the system was under peer review.

What a rescue trial would need to show

A search crew may have to locate someone who is stationary, injured or hidden by vegetation. Purdue’s video begins with the runner already in view. Public accounts of the demonstration give no figures for battery endurance, operating range or the share of attempts in which the drone regained sight of a person after losing them.

A separate prototype from Austria’s Johannes Kepler University Linz offers a benchmark for the initial search. In a 2021 study, researchers reported that an autonomous drone detected 38 of 42 hidden people across 17 forest field experiments. The system combined thermal images from multiple viewpoints to pick out people partly obscured by vegetation and adjusted its flight path to take another look at possible detections.

Purdue has also announced a five-year, $1.5 million agreement with the US Army Research Laboratory for Bera’s team to develop air and ground robots that can map and navigate difficult terrain without GPS. For civilian search teams, field trials would need to measure how much ground the drone can cover on one battery charge, how reliably it can recover a track under different canopy conditions and how accurately it can report a person’s location to crews on the ground.