State of the Art RRT Star Implementation using Deep Neural Networks.

The first GIF Shows the RRT Star Implementation followed by Static Deep RRT Star Implementation

The Second GIF Shows the Dynamic Real Time Deep RRT Star Implementation where the network adapts to changing environments and provides the next step accordingly.

Creates Dataset including 200 start, goal points for each environment and stores it in the csv file in a format Comaptible with the Data Loader.

This piece of code writes the 200 paths for all the 200 environment in output_final.csv file.

After the ouput_final.csv id created there might be instances where the dataset has to be put into the right format. These blocks of code are meant to correct the oytput_final.csv file.

This code includes Contractive AutoEncoder which can be run to obtain encoder weights that will help us encode any given environment from 128 x 128 to 28 x 1.

After the dataset is created in the output_final.csv it can be uploaded in loaded in the Data Loader and the MLP Model can be trained to learn the behavior of RRT-Star Implementation. The model weights obtained are stored in a ".pkl" file.

After training the MLP Model. Any test dataset image with start and goal positions can be specified and the code will provide the path based on the trained model weights.

If you want to visualize the obtained path.

This code is designed to implement Deep RRT Star in Dynamic Environments. If there is any change in the environment. The neural planner takes the new sample space in consideration and provides the next steps accordingly.

You can find the Datasets for the Repository here