Volume 125
Published on October 2025Volume title: Proceedings of CONF-CIAP 2026 Symposium: International Conference on Atomic Magnetometer and Applications
Robust attitude and position control remain critical challenges for consumer drones. This studies evaluates the performance of a Reinforcement Leaning algorithm (PPO) against traditional control algorithms on drone's stability in both computer simulation and real life situations. Reinforcement Learning is trained with common parameters and rewards for small attitude error, lower angular rates, and efficient control effort. Performance were measured across level 0 to 5 wind in simulation and level 0 to 3 in real life experimentation. Results showed that PPO out-performed traditional PID controller in both computer simulation and real life experimentation. PPO showed better stability than PID with reasonable actuation. These findings indicate that PPO can produce more robust, precise control than fixed-gain controllers.
View pdf
