Learning Agile Intruder Interception using Differentiable Quadrotor Dynamics
2026-07-02 • Robotics
Robotics
AI summaryⓘ
The authors developed a way for a drone (called an interceptor) to catch another fast-moving object (the intruder) using only the direction from a single camera, without knowing the exact distance or position. Instead of relying on extra sensors, they used a smart learning method that understands the drone's real flying behavior to improve how it moves quickly and accurately. Their method worked better than older approaches that used simpler flight models by about 30%. This makes it closer to how drones might work in real-world situations with limited information.
deep reinforcement learningcontrol policyquadrotor dynamicspolicy gradientmonocular camerainterceptionpoint mass dynamicspassive sensing3D direction unit vectorautonomous drones
Authors
Michael Anoruo, Xiaoyu Tian, Abhishek Rathod, Timothy Naudet, Thomas Canchola, Eric Sturzinger, Kshitij Goel, Wennie Tabib
Abstract
This paper presents a methodology for learning a control policy to intercept an intruder using the 3D direction unit vector to the intruder and the interceptor state. Prior deep reinforcement learning approaches assume either relative position or distance to the intruder is available, but this information is not readily accessible in real-world applications that employ passive, monocular camera sensors. Instead, we propose a solution that leverages an analytical policy gradient method using differentiable quadrotor dynamics to learn agile interception at speeds up to 10 m/s. The proposed approach outperforms baseline methods that utilize simplified point mass dynamics by an average of 30%.