Optimization of an MPC Based Motion Cueing Algorithm with Deep Deterministic Policy Gradient

In order to optimize the replication of a pilot's sensory dependencies in various flight scenarios within the constrained workspace of a motion platform, adjustments are required to the Model Predictive Control motion cueing algorithm specific to each task scenario. In pursuit of this objective, this study aims to minimize the overall motion cueing error for a given scenario, employing the Deep Deterministic Policy Gradient algorithm from reinforcement learning to train the optimal output weights corresponding to that scenario. The resulting weights, obtained through this methodology, are better suited for specific flight scenarios compared to conventional weight selections. While making efficient use of the motion platform's workspace, these weights provide pilots with a more realistic perception of motion. Moreover, they can be utilized for the construction of multi-model predictive control motion cueing algorithms, facilitating seamless transitions across a broader range of flight scenarios.