Robotic flight meets instinctive adaptation in groundbreaking drone research
The NSF CAREER award recognizes early-career faculty who demonstrate excellence in research and education. It provides five years of stable support to explore transformative ideas.
Conventional drones can manage only rigid objects. Flexible materials, with their unpredictable dynamics, pose complex challenges. "If you try to grab an apple from a branch, the branch pushes back," explains Saldana. "Humans instinctively adjust to these forces. We want robots to do the same."
To address this, Saldana's team is developing a new framework that blends adaptive control systems with reinforcement learning. The goal is to create drones that can react in real time, maintain stability, and improve with experience.
"This means the robot won't need to repeat actions endlessly before mastering them," he says. "It becomes more efficient with each task it completes."
The initiative begins with designing an adaptive controller capable of real-time force compensation without prior material knowledge. This foundation enables reinforcement learning algorithms to help the robot discover optimal control strategies through interaction.
"No one has integrated these two approaches in this way before," says Saldana.
One primary application lies in high-rise construction. Drones could safely position rods and cables, currently handled by workers, improving safety and cutting costs. Other potential uses include deploying plastic sheeting over rooftops during storms and guiding fire hoses during emergencies.
To realize these benefits, developers must overcome the challenge of enabling real-time adaptation in ever-changing environments-a challenge nature has already solved.
"I'm excited to tackle new problems," Saldana says, "and to see how we can give drones the instinctive agility of squirrels."
Research Report:Engineering smarter drones: From nature to complex aerial manipulation
Related Links
SwarmsLab
UAV News - Suppliers and Technology
