trajectory-tracking

This paper introduces a predictive formulation for deep reinforcement learning that augments the state space with future reference horizons to enable anticipatory control for trajectory tracking. Simulation results show significant error reduction, though zero-shot transfer to physical hardware reveals a sim-to-real gap.

This paper presents a two-stage learning pipeline that uses an attention-augmented gated recurrent network to estimate wind from onboard kinematics and then leverages that estimate in a reinforcement learning controller to improve quadrotor trajectory tracking under turbulent winds, reducing tracking error by 48% compared to a baseline.