Purdue University professor Aniket Bera showed an autonomous drone that follows people through dense forests without GPS. Shared on October 24, the video shows the drone predicting a runner's path even when hidden by trees. This technology uses onboard AI to map obstacles and could help future search-and-rescue missions.

Purdue University professor Aniket Bera has demonstrated an autonomous drone that follows a running person through dense forests without GPS or a pre-existing map. Its onboard AI estimates movement, maps obstacles and predicts where a person may go when hidden by trees. The technology could eventually support search-and-rescue operations.

A person runs through a dense forest. A drone follows, weaving between trees. The runner disappears behind a tree trunk, but the drone does not immediately lose track. It predicts where the person may be heading and adjusts its flight to keep up.

A 44-second video shared on LinkedIn on October 24 by Purdue University computer science professor Aniket Bera shows an autonomous drone doing just that, without GPS or a pre-existing map of the forest.

The interesting part is what happens behind the scenes. The drone is not simply following a person on camera. It is estimating its own movement, mapping nearby obstacles and predicting the runner's movement, all while making decisions onboard rather than relying on cloud computing.

HOW DOES IT FOLLOW SOMEONE IT CAN'T SEE?

In his LinkedIn post, Bera explained that the system combines four capabilities.

Visual-inertial odometry helps the drone estimate its own movement by combining camera images with motion-sensor data. Local occupancy mapping builds a picture of nearby obstacles and open spaces as it flies, instead of depending on a map created beforehand.

Then comes dynamic occlusion prediction. When a tree blocks the camera's view of the runner, the system estimates where the person is likely to move based on previous observations.

Finally, perception-aware trajectory optimisation helps the drone choose a route that avoids obstacles while keeping the runner within its camera's field of view.

In simple terms, it has to work out where it is, what is around it, where the person might be going and how to follow without crashing.

WHY THE DRONE DOES NOT NEED THE CLOUD

Bera describes the project as an example of embodied AI, where artificial intelligence processes information and acts in the physical world. The team developed the drone's airframe, software stack and research pipelines at Purdue's IDEAS Lab. Its frame was designed using CAD and 3D-printed carbon fibre.

The entire navigation process runs on embedded hardware. This matters because a drone banking sharply to avoid a thin branch can suddenly change what its camera sees. Waiting for a remote server to process that change could introduce dangerous delays.

Bera says the system must complete the loop from camera images to motor commands in milliseconds, while working within limits on size, weight, power, computing capacity and heat.

He has also clarified that the project uses open-source software tools, but its key new planning and perception algorithms were developed in-house and are under review. A preprint is expected soon.

COULD IT HELP FIND MISSING PEOPLE?

The technology could eventually assist search-and-rescue teams looking for missing hikers or people stranded in difficult terrain. A drone that can navigate without GPS and estimate a person's movement behind vegetation could help search areas that are difficult to cover on foot.

But the video is a research demonstration, not evidence of successful rescue deployments. Its performance in different forests, weather conditions and lighting, along with battery endurance and tracking accuracy, will matter before it can be used in the field.

Bera is an associate professor at Purdue and director of its IDEAS Lab. He completed his BTech in Computer Science and Technology at Jaypee Institute of Information Technology in 2011 and earned a PhD in Computer Science from the University of North Carolina at Chapel Hill in 2017.

Whether the system can be adapted to India's dense forests and disaster-hit regions is a question worth putting to its creator. For now, the demonstration offers a compelling look at what autonomous drones may be able to do when GPS and a clear line of sight are no longer available.