Sampling-based Non-Holonomic Path Generation for Self-driving Cars

Semi-autonomous self driving car technologies are commercially available today, but making them fully autonomous under guaranteed safety is still an open challenge. There are holonomic robot path planning algorithms for generating guaranteed collision-free paths even in very complex but completely known environments. However, it is not feasible to apply these algorithms to non-holonomic cars. In this paper we propose a novel approach to solve this problem in real-time, by generating a series of incremental collision-free path segments from the start to end configuration using an incremental sampling-based planner, such as Rapidly-exploring Random Trees (RRT). The proposed approach employs only local sensor data (point cloud data) for path planning of nonholonomic self driving cars in quasi static unknown environments. The proposed planner generates real-time paths that guarantee safety. The algorithms are extensively testedand validated in popular benchmark simulation environments.