Learning Agile Intruder Interception using Differentiable Quadrotor Dynamics

2026-07-02

AI Summary

This work addresses the practical challenge of high-speed aerial interception under the constraint that only monocular visual observations—providing 3D directional unit vectors without relative position or distance—are available. To this end, the paper proposes a deep reinforcement learning approach grounded in a differentiable quadrotor dynamics model. It is the first to integrate differentiable rigid-body dynamics into a policy gradient framework, combining analytical policy gradients with directional observations to enable agile decision-making using solely visual bearing information and the interceptor’s own state. Evaluated at interception speeds up to 10 m/s, the method achieves an average performance improvement of 30% over baseline approaches employing simplified point-mass dynamics, substantially enhancing both practical applicability and robustness under realistic sensing conditions.

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%.

Problem

• intruder interception
• quadrotor dynamics
• monocular vision
• reinforcement learning
• agile control

Innovation

• differentiable dynamics
• quadrotor interception
• monocular vision
• policy gradient
• agile control