Mixed fidelity training cuts cost of drone swarm teamwork
Calibrate Once, Fly Any Team: Residual-Grounded Low-Fidelity Training for Cooperative Drone Swarms
Multiagent SystemsRobotics
Summary
Training groups of drones to work together usually needs complex and slow simulations that get harder with more drones. The authors created a simpler training method using a basic simulator corrected by data from one drone's short flights, so it works well for many drones without high computing costs. Their method performs almost as well as detailed simulations for large teams but is much faster and avoids many crashes during training. This approach helps build better drone swarms without needing expensive computing resources.
What this means in practice
- •For drone operations teams: Train cooperative drone swarm control policies efficiently without expensive high-fidelity simulations or high crash rates.
- •For robotic software developers: Develop multi-agent control software that scales to large teams using faster low-fidelity simulations corrected by minimal calibration data.
Authors
Maxim Mednikov, Oren Gal
Abstract
Training multi-agent drone-swarm policies directly in high-fidelity (HF) rigid-body physics is accurate but computationally expensive. This cost scales poorly with team size, as each additional agent multiplies contact-resolution complexity and sharply raises the in-simulation crash rate. To address this, we propose a mixed-fidelity training scheme that eliminates HF reinforcement learning entirely. A single shared, decentralized policy is optimized inside a fully-differentiable, JAX-native low-fidelity (LF) point-mass simulator. The simulator is corrected by a small, per-agent bagged residual ensemble fit once, offline, using short calibration flights in the HF simulator. Because calibration requires only one isolated drone, the data collection budget does not compound with team size. Reference trajectories are generated by rolling out an existing LF-only policy and tracked in the HF simulator by a zero-training PD controller. Evaluated across four cooperative drone tasks and team sizes from 3 to 18, the residual-corrected policy outperforms an uncorrected LF baseline in all combinations, and a from-scratch HF policy in 22 of 24 combinations tested. It trails an HF-finetuned policy by a margin that narrows steadily with team size. Ultimately, the proposed method achieves near-equivalent performance at the largest team sizes at a fraction of the computational cost, completely avoiding the high crash rates typical of HF training.