Event
Crazyflow: A GPU-Accelerated Differentiable Drone Simulator
1 reports · 1 sources
Martin Schuck and co-authors propose Crazyflow, a JAX-based differentiable drone simulator that combines high-fidelity dynamics with GPU acceleration, and claim it can scale to millions of environments or swarms of thousands of drones on a single CPU and GPU; it is released as open source for synthetic data generation. The authors claim that real-world experiments show the simulator supports analytic-gradient policy learning, achieving high-precision tracking without domain randomization, and that sampling-based obstacle avoidance exceeds 500 million steps per second; they also threw a physical drone into the air and trained a recovery policy from scratch in only 0.37 seconds, catching it before landing. All of the above results are statements made by the authors in the paper.
Report timeline
- Crazyflow: A JAX-based GPU-accelerated differentiable drone simulator
The authors propose Crazyflow, a differentiable drone simulator built on JAX that combines high-fidelity dynamics with GPU acceleration, scaling to millions of environments or swarms of thousands of drones on a single CPU and GPU. Real-world experiments show that it supports analytical-gradient policy learning, achieving high-accuracy tracking without domain randomization, and sampling-based obstacle avoidance at over 500 million steps per second. The authors also demonstrate in-flight reinforcement learning: after tossing a physical drone into the air, training a recovery policy from scratch took only 0.37 seconds, and the drone was caught before landing.
arXiv cs.RO ·