Uncertainty-aware motion planning for autonomous vehicle: A review

Abstract

This paper reviews a recently developed uncertainty-aware motion planning algorithm vastly applied to autonomous vehicles. Many vehicle manufacturers shifted their focus from improving vehicle energy conversion efficiency to autonomous driving, aiming to bring a better and more relaxed driving experience to drivers. However, many past motion planning algorithms used for autonomous driving were immature, so many errors were reported. These errors may put human drivers in life-threatening danger. Consisting of two connected systems supported by a well-trained graph neural network, the uncertainty-aware motion planning algorithm uses two related sub-systems to predict the motion of surrounding object and make necessary maneuvers accordingly. Using evidence from many research papers, an uncertainty-aware motion algorithm is an efficient and safe solution to insufficient consideration of the surrounding environment of vehicles. Even though its ability is primarily limited by the accuracy of sensors and the complexity of background, the unique advantage of this algorithm gives an alternative direction to the development of algorithms in autonomous vehicles.