This work was accepted at the International Conference on Computer Vision (ICCV) 2025 Honolulu, Hawaii, USA.

This work introduces a self-supervised neuro-analytical, cost efficient, model for visual-based quadrotor control in which a small 1.7M parameters student ConvNet learns automatically from an analytical teacher, an improved image-based visual servoing (IBVS) controller. Our IBVS system solves numerical instabilities by reducing the classical visual servoing equations and enabling efficient stable image feature detection.

Our key contributions include: 

• An analytical IBVS teacher that solves numerical instabilities inherent in classical approaches

• A two-stage segmentation pipeline combining YOLOv11 with a U-Net-based mask splitter for robust anterior-posterior vehicle segmentation to correctly estimate the orientation of the target 

• An efficient knowledge distillation dual-path system, which transfers geometric visual servoing capabilities from the analytical IBVS teacher to a compact and small student neural network that outperforms the teacher, while being suitable for real-time onboard deployment.

This work was accepted at the European Robotics Forum (ERF) 2026 Stavanger, Norway.

We introduce a vision-language-guided framework where a frozen model generates pseudo-labels to train a lightweight YOLOv8 student for markerless drone detection. Coupled with a depth-free IBVS controller operating solely on 2D image features, our system enables robust, real-time indoor UAV tracking with minimal manual supervision. Validated in classroom flights, it performs reliably across varied trajectories and supports deployment for surveillance, inspection, and, automation.

This work addresses the challenge of enhancing small UAV self-flying capabilities by enabling them to perform complex tasks using only camera sensors, eliminating the need for GPS and expensive additional equipment such as LiDAR. The proposed system integrates semantic segmentation and monocular relative depth prediction to improve obstacle avoidance, scene navigation, base recognition, and autonomous return capabilities. The semantic segmentation together with the relative depth, camera height and parameters are used to obtain an adaptive metric depth used for improved decision-making.

More information after publication.

This system enables intuitive drone operation through natural hand and body gestures, eliminating the need for traditional remote controllers. Users can command the drone to follow their hand or send movement commands. Rotational movements are executed by rotating the hand in the desired direction. The system includes safety features such as emergency stop gestures and automatic hover when no clear command is detected. This approach makes the drone operation more accessible and intuitive, particularly useful for content creation, surveillance applications, and situations where traditional controllers are impractical.