MinNav lets tiny drones navigate obstacles using only optical flow
70% success rate navigating unknown gaps with a single camera
MinNav, developed by Aniket Patil, Mandeep Singh, Uday Girish Maradana, and Nitin J. Sanket, is a minimalist navigation system for tiny aerial robots that relies solely on optical flow—motion of pixels between frames—and its uncertainty. The system requires no prior knowledge of the environment, making it capable of handling static obstacles, dynamic obstacles, and unknown-shaped gaps in real time. By exploiting the robot's activeness (moving around to explore), MinNav improves obstacle detection and navigation success.
The approach was evaluated in diverse real-world environments and achieved a 70% overall success rate. According to the authors, this is the first monocular camera solution to tackle all those navigation cases without prior information. It performs comparably to depth-based methods but with orders of magnitude less computation, enabling onboard execution on tiny aerial robots. The work has been accepted at ICRA 2026, and code, datasets, and supplementary materials are publicly available.
- Uses optical flow and its uncertainty for navigation, eliminating the need for depth sensors or prior knowledge
- Achieved 70% success rate in real-world tests with static/dynamic obstacles and unknown-shaped gaps
- First monocular-only method to handle all these cases, matching depth-based performance with far lower computational cost
Why It Matters
Enables cheap, tiny drones to autonomously navigate complex environments, unlocking applications in search-and-rescue, inspection, and surveillance.