Spiking Neural Networks for High-Speed Continuous Quadcopter Control Using Proximal Policy Optimization

Abstract

We present the first fully spiking actor-critic neural network policy, trained via Proximal Policy Optimization (PPO), for continuous control of a high-speed quadcopter in a gate-based navigation task. The spiking neural network (SNN) controller uses Leaky Integrate-and-Fire neurons with surrogate gradient training and spike-rate decoding over multiple integration cycles. The SNN is benchmarked against a comparable artificial neural network (ANN) in both simulation and real-world flight. Despite training to the same reward level, the SNN achieves superior simulation performance with higher episode rewards, greater robustness, and reduced crash rates. In 12-second real-world trials, the SNN also outperforms the ANN, attaining higher average reward (70.63 vs 59.77), greater mean velocity (7.94 vs 6.99 m/s), and more gates cleared (46.33 vs 40.67). Analysis of the spike integration cycle count reveals a trade-off: lower counts reduce control output resolution and hinder learning, while higher counts improve smoothness but increase inference latency. Moderate cycle counts (five or eight) provide the best balance of high rewards, smooth outputs, and low overhead. These findings demonstrate that SNN-based policies can outperform conventional ANN controllers in high-speed, agile robotic tasks, representing a key step toward neuromorphic control in embedded autonomous systems.