Learning Collision Checking Policies

Official repository for evaluating collision checking policies that use bayesian active learning techniques

Related Publications

1. Near-Optimal Edge Evaluation in Explicit Generalized Binomial Graphs (NIPS 2017)
2. Bayesian Active Edge Evaluation on Expensive Graphs (IJCAI 2018)

Datasets

The repository containing datasets is graph_collision_checking_dataset

Setup

1. Clone the repository and the datasets folder
2. cd matlab_learning_collision_checking
3. Edit init_setup.m to add the global path to the datasets folder by editing setenv('collision_checking_dataset_folder','/path/to/data')
4. Run install_dependencies.m
5. Run init_setup.m

Executing the algorithms

Run the file src/benchmark_coll_check_policy.m to execute the algorithms in the paper on the datasets

Creating 2D datasets

1. Go to dataset_processing/2D_dataset_creation/
2. Run any one of the scripts from example_environments/ to generate a set of environments corresponding to some world distribution. You may have to create a set of empty folders for the scripts to save stuff in.
3. Run create_graph.m. This will create a 2D RGG, start and goal and save this.
4. Run collision_check_graph.m. This will collision check the graph on a given dataset.