I'm curious about the technical challenges of autonomous indoor drones: what sensor combinations (vision, LiDAR, ultrasonic) are most effective for obstacle avoidance? And how can SLAM algorithms adapt to changing environments without GPS? What trade-offs between accuracy, energy consumption, and processing time do you find most relevant? I'm interested in your practical experiences and theoretical references.
Current limits of indoor navigation for autonomous drones?
👁️ 66 views💬 1 replies❤️ 0 likes
1 Replies
Same here, I also had this problem: I tested several configurations on a small indoor quadcopter, and the vision + ultrasonic combo turned out to be the most robust. The RGB-D camera (or stereo vision) provides a good point density for SLAM, while ultrasonic sensors fill in areas where texture is low (white walls, smooth floors). The 2D lidar adds extra precision, especially for corners and very close obstacles, but it consumes more power and generates a heavy data stream to process in real time.
On the algorithm side, I preferred a lightweight version of RTAB-Map that integrates both visual loop closures and distance measurements from ultrasonic sensors. In changing environments (moving furniture), an EKF-SLAM filter helps merge the two sources and reduce drift without GPS. The trade-off I found most effective: accepting a 10 cm accuracy to keep power consumption under 5W and processing time below 30 ms per frame. If you need higher precision, a 3D lidar is the way to go, but you’ll have to sacrifice either battery life or refresh rate. Useful references: “Vision-Based Indoor SLAM” by Engel et al., 2014, and Huang et al.’s (2021) paper on lidar-ultrasonic fusion for confined-space navigation.