Single aerial drones help people in many ways, such as assisting in search and rescue missions. Drone swarms can be similarly handy, accomplishing tasks including surveying endangered wildlife in harsh and remote environments. But there are some environments where flying robotic teams still struggle to go. For example, crowded airspaces like thick woods can challenge swarms trying to plan a flight path.
Now, a flight path planning method published in the May 4 issue of Science Robotics offers a new way for drone swarms to successfully fly in crowded, unfamiliar airspaces. When tested, the system helped 10 small autonomous flying drones efficiently pick the best way to fly through a cluttered bamboo forest. Using the new approach, the palm-sized drones also maintained formation, avoided collisions, and tracked a human in unfamiliar environments.
"How to build a robotic swarm is always the ultimate target of many robotic researchers around the world," said Xin Zhou, a Ph.D. candidate at Zhejiang University and lead author of the study. "Our aerial robots are equipped with stereo cameras and onboard computers, which allow them to precisely understand the environments and plan nice trajectories. Also, they share information among the group and cooperate with others, to mimic the swarm flight of birds in challenging natural environments."
Airspace, One of the Final Frontiers
Traditionally, autonomous aerial robots calculate their flight trajectory under specific space and time constraints. This is harder for robotic swarms to do because they have to synthesize individual paths to build a collective path plan. Swarm path planning also needs to incorporate several factors to be truly useful in search and rescue missions.
"Imagine a forest search and rescue scenario where we need to search for missing persons," said Zhou. "Time is life, and the swarm needs to plan smooth and fast trajectories to cover as much area as possible within a limited time. This is trajectory optimality."
Other elements for effective swarm path planning include extensibility, or the ability to assess and accommodate multiple goals like reducing the need for repeated searches and avoiding collisions at the same time. There's also economical computing, which means how much effort it takes to deploy path planning algorithms on the robots' onboard computers, and miniature size, which allows for navigation in narrow and restricted areas.
With these factors in mind, Zhou and his colleagues designed a trajectory planning method that processed data from embedded computers on 10 micro-drones to help those drones swarm in four unfamiliar environments.
Testing the Swarm
To test whether their small aerial drones could use the new trajectory planning method effectively, the researchers put the swarm through a variety of scenarios.
"Each scenario is designed to verify the characteristics of our system in different aspects," said Fei Gao, an assistant professor at Zhejiang University and corresponding author of the study.
For the first test, the robotic swarm used the planning system to fly through dense bamboo forest. The drones adjusted their trajectories in response to obstacles like fallen bamboo, skillfully passing through narrow openings between stalks roughly less than 30 centimeters wide. This test showed how that system incorporated trajectory optimality and miniature size.
Next, the researchers evaluated the drone swarm's ability to stay in formation, demonstrating extensibility by assigning the swarm a moving shape while navigating an unknown airspace. During this test, the drones kept their overall formation while flying through standing bushes, trees, and two human-made pillars.
Following the formation test, the robotic swarm zoomed through an aerial traffic scenario, where each drone's flight path was pre-designed to intersect with others. The drones had to adapt in the moment to avoid collisions. Zhou, Gao, and their colleagues had the swarm track and film a single person, testing the system's economical computing and extensibility.
"Only economical computing algorithms allow sufficient computational resources to be reserved for adding more sophisticated visual tracking algorithms. And tracking itself is a new task requirement here for our planning scheme," said Gao.
All of the tests indicated the new path planning system can enable autonomous aerial robotic swarms to coordinate and achieve previously challenging tasks in crowded airspace.
"We believe that our system and algorithm are useful in many tasks that require autonomy and coordination in the air," said Gao. "We think that search and rescue teams, animal and plant researchers, and even each of us expecting aerial traffic in the future, could gain inspiration from this work."
[Credit for associated image: Reprinted from Zhou et al., Sci. Robot. 7, eabm5954 (2022)]