Motion Planning of Autonomous Vehicles under Dynamic Traffic Environment in Intersections Using Probabilistic Rapidly Exploring Random Tree

Citations Over Time

Abstract

<div>In motion planning of autonomous vehicles, non-signalized intersections pose challenges due to a variety of traffic flows. Common motion planning algorithms use the current environmental information to find an optimal path that satisfies traffic safety and efficiency. Because of the non-signalized intersection dynamics, the algorithms need to iteratively generate a path while avoiding collision with other obstacles. Traditional grid-based planning algorithms present an enormous computational burden for real-time implementation. Meanwhile, sample-based algorithms like Rapidly exploring Random Trees (RRT) could be used for local motion planning to determine possible safe paths and quickly reselect alternative paths towards the goal. However, near an intersection, estimating another vehicle’s dynamic state and avoiding collision through standard RRT can be cumbersome. Hence, this article introduces a novel Gaussian Processes Regression (GPR) technique to predict the future location of an involving vehicle in a space that combines with the standard RRT algorithm for intersection traffic motion planning. The predicted location information is then used in both sampling and steering paths of the ego vehicle to ensure that the ineffectual sampled location can be avoided, and path segments are collision-free for all time instances. Benchmarking the proposed algorithm with the standard RRT demonstrated an increase in the success rate of generating a safe path and a significant decrease in execution time due to the sampling bias towards the destination. The algorithm is further evaluated through simulation in a four-way intersection when another vehicle passes through the intersection. The proposed method can be a precursor towards a novel field of research for combining probabilistic maps for ineffectual sampled locations with planning methods like RRT* and informed RRT* and can be applied to multi-vehicle motion planning too.</div>

Related Papers

• → An obstacle-based rapidly-exploring random tree(2006)203 cited
• → Rapidly-exploring random tree based memory efficient motion planning(2013)117 cited
• → Extension of the Rapidly Exploring Random Tree Algorithm with Key Configurations for Nonholonomic Motion Planning(2006)17 cited
• → A RRT*FN Based Path Replanning Algorithm(2019)11 cited
• Bidirectional Rapidly-exploring Random Tree Algorithm of Ship Assembly/Disassembly Path Planning Based on Motion Constraint Channels(2010)