An easy-to-use yet efficient motion planning library with focus on providing a modular framework for integrating planning algorithms with machine learning frameworks. The library allows for loading planning environments from png files and currently supports lattice based planning with custom heuristics and cost functions.

The search based planning pipeline looks as follows:

1. Numpy
2. Matplotlib
3. dubins

Once you have the external dependencies installed,

• minimal_astar_example.py: A simple example that loads an environment and runs a 2D astar planner with heuristic weight 1. The details of the parameters used are as follows:

• astar_2d_benchmark_example.py: This script takes as input the relative location of different planning problem databases and number of environments in each database. It then runs a 2d astar planner with changing heuristic weights and stores the results in the benchmark_results folder

We provide results of benchmarks run on different motion planning datasets. We run A-star with varying the weight on the heuristic function from 0 to 91. A heuristic weight of 0 makes the algorithm equivalent to Dijkstra's Algorithm whereas a higher weight makes the algorithm more greedy. The results reported are average number of expansions on a set of 100 environments for each database.

• A-star 2D Benchmark Results

  Parameters:  graph size                    = 200x200
               connectivity                  = eight-connected
               cost                          = path length
               heuristic                     = Euclidean distance to goal
               collision checking resolution = 1m