Dynamic 2D Environment Generator

This is a 2D environment generator useful for random generation of random layouts in 2D. The implementation has been inspired by the environment description as specified in Appendix G of RoCUS. A few application where such an environment representation is useful are: environments for robot 2D navigation, obstacle grids, random meshes, maze generation, etc.

• Dynamic 2D Environment Generator
  • Environment Description
  • Parameters
  • Requirements
    • To use the environment description from the environment.py file, the following dependencies are required:
    • Additional requirements to generate environments in bulk using the main.py file:
    • The source has been tested with the following versions:
  • Sample Environments Generated
    • Parameters Used
      • Offset 0:
      • Offset 20:
      • Offset 40:
      • Offset 60:
      • Offset 80:
  • Generated Environments

Environment Description

The environment is the area defined in the domain of (xmin, xmax) × (ymin, ymax) along with starting point (xstart, ystart) and ending point (xgoal, ygoal). The environment representation is a summation of radial basis function (RBF) kernels centered "obstacle points" that are randomly generated based on a seed value. Specifically, given n obstacle points $p_{1}, p{2}, ..., p{n} ∈ R^2$, the environment is defined at each point in the domain R^2 (within limits) as:

$$e(P) = \sum_{i=1}^{n} \exp(-\gamma || P - P_{i} ||_{2}^2)$$

A point P in the limits of the environment (xmin, xmax) × (ymin, ymax) is an obstacle if $e(P) > \eta$. Note: the environment is a continuous space and not discretized. The environment needs a resolution if it is to be discretized into a configuration space.

Parameters

- environment_range: (min limit, max limit) : the limits of the square environment.

- obstacle_point_range: (min limit, max limit) : points in the grid that can be obstacles.

- start_point: (start x, start y) : starting point in the environment.

- end_point: (end x, end y) : ending point in the enviropnment.

- obstacle_points: int ∈ (0, inf) : number of "obstacle points" for the kernel.

- eta: float : threshold (min weight) for classifying point as an obstacle.

- grid_resolution: int : resolution of grid for discretization.

- seed: int : a seed value for the psrng to generate the "obstacle points" kernels.

- gamma: int : weight for the RBF kernel.

Requirements

To use the environment description from the environment.py file, the following dependencies are required:

numpy
matplotlib

Additional requirements to generate environments in bulk using the main.py file:

tqdm

Execute the main file using the command python main.py from src as the working directory.
Note: make sure that the output file path exits.

The source has been tested with the following versions:

matplotlib==3.1.2
numpy==1.17.4
tqdm==4.64.1

Install all requirements using pip install -r requirements.txt from project source.

Sample Environments Generated

Parameters Used

environment_range = (-1.2, 1.2)
obstacle_point_range = (-0.7, 0.7)
start_point = (-1, -1)
end_point = (1, 1)
obstacle_points = 15
eta = 0.9
grid_resolution = 500
seed = 1337 + offset (offset is set as per the image number)
gamma = 25

Offset 0:

Offset 20:

Offset 40:

Offset 60:

Offset 80:

Generated Environments

An archive of 5000+ environment layouts generated using the parameters listed above can be found at this link.