A Safe Decision Making Framework for Automated Vehicle Navigation among Human Drivers

Abstract

In the near future, automated vehicles (AVs) will have to interact closely with Human-driven vehicles (HDVs). This work proposes an integrated decision-making framework that considers HDVs motion, a feasibility check, and planning. A learning-based encoder-decoder Long Short-Term Memory is used for HDV motion prediction. An error ellipse is used to capture the uncertainty from the learning-based model. A feasibility check is carried out to confirm the existence of a lane change trajectory from the given target vehicle's future position. The results from the feasibility check decide the action of AV. This work uses a lower-order parametric curve for path planning combined with an efficient trapezoidal acceleration-based velocity planner. Simulation results show that the proposed method guarantees a collision-free path for a lane changing scenario, given the lead vehicle position.