Combining Stochastic and Scenario Model Predictive Control to Handle Target Vehicle Uncertainty in an Autonomous Driving Highway Scenario

Abstract

Autonomous vehicles face the challenge of providing safe transportation while efficiently maneuvering in an uncertain environment. Considering surrounding vehicles, two types of uncertainties occur: multiple future maneuvers are possible, and within these maneuvers the vehicle can vary from the predicted ideal maneuver path. Focusing on only one of these uncertainties can either lead to neglecting potential risks or result in overly conservative motion planning. Here, we suggest a Stochastic Model Predictive Control strategy that tackles the possibility of multiple future maneuvers of surrounding vehicles, while also considering uncertainty within the execution of these predicted maneuvers. The proposed control method is a combination of Scenario Model Predictive Control to cope with multiple predicted maneuvers of other vehicles, and Stochastic Model Predictive Control using chance-constraints to take into account vehicle deviations from the predicted maneuver trajectories of the respective maneuver. Adjustable risk parameters permit a violation of safety constraints up to a desired probability, allowing a trade-off between risk and performance. A simulation of a two-lane scenario demonstrates the effectiveness of our method.