Reactive Online Motion Re-Planning for Crash Mitigation in Autonomous Vehicles Using Bezier Curve Optimization

Unexpected driving behavior and reckless driving on roads have drastically increased the number of accidents in the past few years. This could be attributed to drinking and driving, getting sleepy or losing focus on the road while driving, which could lead to crash situations with other vehicles. Reacting quickly to avoid any dynamic obstacle on the road is necessary for crash avoidance. In most of the current collision avoidance systems (CAS), braking or speed adjustment are the most common methods employed. But in emergency situations, braking will not always be able to avoid a crash. Implementing a maneuver around the dynamic obstacle is a method employed by current Advanced Driver Assistance Systems (ADAS). This maneuvering approach in high speed scenarios poses many challenges like vehicle stability, maneuver generation for least crash severity. The existing research in path planning methods for dynamic obstacle avoidance have not addressed crash severity, computational complexity and vehicle dynamic constraints simultaneously.

This paper considers an emergency crash situation with an oncoming obstacle vehicle where braking is not feasible to avoid the crash. A crash mitigation strategy is proposed for this situation in autonomous vehicles using reactive path replanning. Cubic Bezier curves are used to generate the avoidance maneuver around the dynamic obstacle/vehicle. The distinct advantage of using Bezier curves method lies in less computational complexity and faster manipulation of the generated trajectory for obstacle avoidance as compared to the existing path planning approaches. The problem is converted to a non-linear optimization problem, where  constraints like crash severity and vehicle dynamics are considered for solving the optimization problem. The overall collision avoidance strategy involves three steps: (1) collision checking, (2) trajectory replanning and (3) solving the optimization problem. The given optimization problem is solved using Quadratic Programming and Hildereth’s algorithm. Two assumptions are made in this study: (1) The vehicle is equipped with appropriate sensors to track the oncoming dynamic vehicle’s position (2) The dynamic obstacle’s motion prediction has been done and its path can be represented by a Bezier curve. This allows the vector difference between the generated maneuver and the predicted trajectory to be represented by a Bezier Curve. This allows properties of the Bezier curve to be used for collision checking and trajectory generation. 

This proposed collision avoidance approach is validated for two on-road scenarios using simulation results, where an oncoming obstacle car is considered. It is validated that the Bezier curve approach is efficient and computationally fast for avoiding a dynamic obstacle. Moreover, in an unavoidable crash situation, it is able to reduce the crash severity using an appropriate lateral steering control and feasible maneuver.