Motion Prediction for Teleoperating Autonomous Vehicles using a PID Control Model

Teleoperating autonomous vehicles is challenging due to latency and bandwidth constraints. In order to increase operator safety and situation awareness, techniques similar to motion planning for control of autonomous cars in dynamic environments have been adapted for aerial vehicles in this study. An overview of a novel concept based on reconstruction of the environment, user handling, and predictive modeling will be given. The working principle of predictive motion for teleoperating vehicles is explained and key metrics are introduced to compare changes of model parameters. A proportional-integral-derivative (PID) control model has been developed and integrated into the concept. The concept has been evaluated based on flight simulations as well as with actual test flights. The sensitivity of the PID parameters and the impact of the correct estimation of the predicted latency were investigated. The concept has been successfully been demonstrated with a DJI M600 hexacopter. The analysis indicates a high sensitivity for the P-component and low sensitivity for I and D components for an accurate prediction. Latency analysis shows that underestimation of the real latency does not have as high an impact as overestimating it and that the model fits best for latencies below 250 ms. Furthermore, the implemented model lacks the prediction accuracy in the acceleration phase and a representative inertial model. The here presented model is a novel approach to handle the predicted motion of teleoperated vehicles and shows promising results in accuracy and parameter sensitivity.