Skip to main content

New Algorithm Optimizes Data Collection by Drones on Antarctica

A path-optimizing algorithm helped drones survey the Cape Crozier penguins in two hours compared to the usual three days. | Shah et al., Sci. Robot. 5, eabc3000 (2020)

A team of drones surveyed two Adélie penguin colonies in Antarctica in record time, thanks to a new algorithm that optimizes paths for the drones, such that they cover large stretches of land on a single trip with minimal backtracking.

Using the path-planning algorithm, scientists collected data on one of the largest Adélie penguin colonies in the world, Cape Crozier, in three hours, compared with an average of two days using manually operated drones.

Their study, presented in the October 28 issue of Science Robotics, demonstrates one way autonomous robots can advance conservation efforts, by collecting data faster and more efficiently than humans in extreme environments. Accurate and timely evaluations of wildlife populations and ecosystem changes are especially critical in environments like Antarctica that are threatened by a changing climate.

The new algorithm can also come in handy for directing drones over disaster sites, wildfire-prone forests, or any environments that are difficult for humans to safely access and where time is of the essence, noted Kunal Shah, a Ph.D. candidate of mechanical engineering at Stanford University and lead author of this work.

"In these scenarios getting an accurate and complete understanding of the situation quickly could really benefit responders," said Shah. "Unlike humans, drones don't need to navigate landslides, downed power lines, and other ground hazards."

However, even drones face challenges when navigating the rapidly changing weather and constant shuffling of penguins in Antarctica. Drones with fixed wings — which often look like predatory birds — can scare away the penguins and are difficult to fly at lower altitudes. Drones with rotating blades — the kind used in this study — are less disruptive and can fly closer to the penguins, but they experience constraints on battery life and imaging resolution in such frigid and extreme conditions.

To address these limitations, Shah and his colleagues wanted to find a way for drones to conduct wildlife surveys as quickly and efficiently as possible. To test their algorithm, they targeted Adélie penguin colonies of Cape Royds and Cape Crozier on Ross Island, where penguins have been studied for more than a century. With more than 300,000 nests spreading over an area of 2 square kilometers or 0.77 square miles, Cape Crozier also harbors one of the largest Adélie penguin colonies in the world.

"These are large areas and figuring out the best path to fly is not always straightforward for one drone, let alone a whole team," said Shah.

The new algorithm tackled this problem by first dividing the survey area into a grid. It then selected winding paths that meet all grid points, starting small and adding more connections to each path, resulting in the most efficient set of paths for the drone team. In this way, work could be simultaneously split among the drones, ensuring that no paths overlapped and that areas were not passed over more than once.

"Robotic systems like ours allow field ecologists and conservationists to get population data quickly and with low effort," said Shah. "I think a big contribution of path planners like this, moving forward, is freeing up researchers from having to do important, but repetitive tasks, such as manually flying drones or even manually counting things."

Accounting for constraints in battery life, altitude, and airspace, the algorithm also planned safe return trajectories for drones that had reached their path's end, or those that met technical difficulties in the middle of a mission. It chose return paths that were different from departure paths, to ensure that the drones exploited every second of airtime to cover as much ground as possible.

Previous path planning methods for drones often don't consider the return trip, noted Shah, and as a result, batteries can drain before the mission is complete. And when a drone is travelling over kilometers of ice, snow, or water, a dead battery can be a big problem.

Using the algorithm, four multirotor drones took more than 2,000 images of penguin colonies during each survey, which were later stitched into a composite visual of the area. In collaboration with ecologists, Shah and his team plan to use these images to assess nesting habitats, nest density, and breeding success for the two penguin colonies.

"A big part of conservation is data gathering," said Shah. "Unfortunately, animals are not going to mail in census data." His drone-led system highlights the benefit of using robots in conservation research and raises the question of what else could benefit from automation in field studies.

"Can we push the limits of robot autonomy further to allow agents to decide where to go to collect targeted data, for instance, about a particular penguin colony once it is found?" wrote Marija Popović, a postdoctoral researcher of robotics at Imperial College London, in a piece discussing the study.

Shah and his team are contemplating this and other theoretical questions to improve their path-planning algorithm and are looking to survey different kinds of field sites, including lakes, mountains, and agricultural land.

"Automation has had a large impact on human quality of life, and I'm excited to apply that to environmental problems and use drones for social and environmental good," said Shah.