icra2018/learnedsamplingdistributions repository overview

⁠Learning Sampling Distributions

The repository follows along with the paper "Learning Sampling Distributions for Robot Motion Planning" by Brian Ichter, James Harrison, and Marco Pavone of Stanford's Autonomous Systems Lab, submitted to ICRA 2018.

The iPython notebook learns a sampling distribution for a narrow passage problem from the included data set. The distribution is computed through a conditional variational autoencoder, where samples are generated by sampling the latent space conditioned on the problem's initial state, goal state, and an occupancy grid of the obstacles.

Planning code to run the sample sets or generate data can be found at https://github.com/StanfordASL/GMT⁠.

⁠How to Run with Docker

⁠Linux

⁠Prerequisites

• NVIDIA GPU with CUDA support
• nvidia-docker 2.0⁠

Tested on Ubuntu 16.04.6 with Docker 18.06.1-ce, GPU GeForce 940M, NVIDIA Driver version 410.48.

1. Open a terminal and run the command:

nvidia-docker run --rm -p 8888:8888 icra2018/learnedsamplingdistributions:latest

1. Run a web browser and open the link: http://localhost:8888/lab/tree/README.ipynb⁠