**6. Conclusion**

generated rapidly exploring random tree just follows the front vehicle with a proper distance. Finally, when takeover is possible, the sampling domain will be expanded to include suitable

**Figure 7.** Simulation results in different scenarios. The initial and final positions of the vehicle are shown by yellow

The average results of the performance of the proposed planner as well as the performances

**Scenario Variable RRT RRT\* Proposed algorithm**

Average optimality (%) 63.4 93.8 93.9 Average runtime (s) 87.6 114.0 48.3

Average optimality (%) 50.7 91.4 91.4 Average runtime (s) 134.2 168.9 93.5

Average optimality (%) 68.6 93.5 93.4 Average runtime (s) 73.5 108.6 46.4

Free drive Average number of samples 314.2 203.7 138.8

Takeover Average number of samples 627.2 808.5 316.4

Follow Average number of samples 223.1 418.5 108.8

**Table 1.** Simulation average results for the proposed algorithm, RRT and RRT\* planners.

space.

34 Autonomous Vehicle

and green, respectively.

of RRT and RRT\* planners are shown in **Table 1**.

In this chapter, a novel sampling-based navigation architecture has been introduced that is capable of driving autonomous vehicles in crowded environments. The proposed planner utilizes the optimal behaviour of the RRT\* algorithm combined with the low runtime require‐ ments of low-dispersion sampling-based motion planning. Furthermore, a novel segmentation procedure is introduced which differentiates between valid and tabu segments of the sampling domain in different situations.

Simulation results show the robust performance of this planner in different scenarios such as following the front car and takeover. This method also outperforms the classical RRT and RRT\* planners in terms of runtime and the size of the generated tree while maintaining the same optimality rate as RRT\*.
