1: \begin{abstract}

2: 

3:   Path planning plays a crucial role in various autonomy applications,

4:   and RRT* is one of the leading solutions in this field. In this

5:   paper, we propose the utilization of vertex-based networks to

6:   enhance the sampling process of RRT*, leading to more efficient path

7:   planning.

8: 

9:   Our approach focuses on critical vertices along the optimal paths,

10:   which provide essential yet sparser abstractions of the paths. We

11:   employ focal loss to address the associated data imbalance issue,

12:   and explore different masking configurations to determine practical

13:   tradeoffs in system performance. Through experiments conducted on

14:   randomly generated floor maps, our solutions demonstrate significant

15:   speed improvements, achieving over a 400\% enhancement compared to the

16:   baseline model.

17:  

18:   

19: %In this article, we present a novel approach to improve the learned

20: %heuristic for path planning algorithms. Rather than using the entire

21: %optimal path line segment as the objective, we focus solely on the

22: %turning points of the optimal path. This modification significantly

23: %enhances the speed of the path planning algorithms. To address the

24: %imbalanced data problem caused by the vertex objective, we employ the

25: %Focal Loss technique in our deep learning model. Additionally, we

26: %introduce a mask to the neural network sampler, which boosts the

27: %sampling probability of the vertices in the optimal path. This further

28: %accelerates the convergence speed to the optimal path.

29: 

30: %Overall, our approach provides a more efficient method for learning

31: %heuristics, which is important for future machine learning

32: %applications in path planning problems.

33: 

34: \end{abstract}

35: