UWB Meets Crazyflow: Simulating Degraded Feedback at Scale for Aerial Robotics

2026-10-06

AI Summary

This study addresses the training inefficiency and absence of real-world degradation feedback in aerial robot simulation caused by the decoupling of physics and control. To this end, it proposes Crazyflow, a JAX-based differentiable simulator. By leveraging XLA just-in-time compilation to unify the physics and control computation graphs, Crazyflow achieves end-to-end differentiability. Furthermore, its modular architecture integrates high-fidelity UWB and IMU sensor noise models, enabling large-scale parallel reinforcement learning. Experimental results demonstrate that Crazyflow delivers order-of-magnitude simulation acceleration, completing policy training and deployment within seconds while efficiently evaluating control performance under realistic degradation conditions. Consequently, this work provides a critical foundational tool for advancing next-generation aerial robotics research.

Abstract

In this work, we introduce Crazyflow, an accurate, differentiable simulator built on JAX. By leveraging jit compilation via XLA, Crazyflow unifies physics and control into a single differentiable computation graph, enabling massive parallelization on accelerated hardware without sacrificing modeling accuracy. This architecture achieves order-of-magnitude speedups over existing baselines, capable of training deployable reinforcement learning agents in seconds. To highlight its highly modular design, we demonstrate how easily Crazyflow can be extended by integrating a complete, high-fidelity Ultra-Wideband (UWB) and Inertial Measurement Unit (IMU) simulation pipeline coupled with a full-state Extended Kalman Filter (EKF). This capability allows for massive parallel controller evaluation under realistic, degraded state feedback with minimal impact on GPU throughput. By combining speed, accuracy, and extensibility, Crazyflow serves as a foundational tool for the next generation of aerial robotics research.

Problem

• Aerial Robotics
• Differentiable Simulation
• Reinforcement Learning
• Degraded Feedback
• Massive Parallelization

Innovation

• Differentiable Simulator
• JAX
• Reinforcement Learning
• Ultra-Wideband (UWB)
• Aerial Robotics