发表机构
Technical University of Munich; University of Toronto; Simon Fraser University（慕尼黑工业大学; 多伦多大学; 西蒙菲莎大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文介绍基于 JAX 的可微分模拟器 Crazyflow，通过 XLA 编译实现大规模并行，显著加速训练，并集成 UWB/IMU 与 EKF 流水线，支持退化反馈下的高效控制器评估，为空中机器人研究提供基础工具。
AI 中文摘要
在这项工作中，我们介绍了 Crazyflow，一个基于 JAX 构建的精确、可微分模拟器。通过利用 XLA 的 jit 编译，Crazyflow 将物理和控制统一到单个可微分计算图中，从而在加速硬件上实现大规模并行化，同时不牺牲建模精度。该架构相比现有基线实现了数量级的加速，能够在数秒内训练可部署的强化学习智能体。为了突出其高度模块化的设计，我们展示了通过集成一个完整的高保真超宽带（UWB）和惯性测量单元（IMU）仿真流水线，并耦合全状态扩展卡尔曼滤波器（EKF），可以轻松扩展 Crazyflow。这一能力允许在现实的退化状态反馈下进行大规模并行控制器评估，同时对 GPU 吞吐量的影响极小。通过结合速度、精度和可扩展性，Crazyflow 为下一代空中机器人研究奠定了基石。
英文摘要
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.
CommentsAccepted to 1st Workshop on Robot Meets GNSS and Ranging for Seamless Autonomy @ ICRA 2026
