Vector Field Augmented Differentiable Policy Learning for Vision-Based Drone Racing

TL;DR

This paper introduces DiffRacing, a vector field-augmented differentiable policy learning framework that improves autonomous drone racing by enhancing gradient signals and enabling efficient sim-to-real transfer.

Contribution

It presents a novel integration of vector fields and differentiable losses into policy learning, specifically tailored for high-speed drone racing tasks.

Findings

Achieves superior sample efficiency and faster convergence in simulations.

Demonstrates robust real-world drone racing performance.

Enables effective sim-to-real transfer without explicit system identification.

Abstract

Autonomous drone racing in complex environments requires agile, high-speed flight while maintaining reliable obstacle avoidance. Differentiable-physics-based policy learning has recently demonstrated high sample efficiency and remarkable performance across various tasks, including agile drone flight and quadruped locomotion. However, applying such methods to drone racing remains difficult, as key objective like gate traversal are inherently hard to express as smooth, differentiable losses. To address these challenges, we propose DiffRacing, a novel vector field-augmented differentiable policy learning framework. DiffRacing integrates differentiable losses and vector fields into the training process to provide continuous and stable gradient signals, balancing obstacle avoidance and high-speed gate traversal. In addition, a differentiable Delta Action Model compensates for dynamics mismatch, enabling efficient sim-to-real transfer without explicit system identification. Extensive simulation and real-world experiments demonstrate that DiffRacing achieves superior sample efficiency, faster convergence, and robust flight performance, thereby demonstrating that vector fields can augment traditional gradient-based policy learning with a task-specific geometric prior.