In this work, we use Generative Adversarial Network (GAN) to estimate the distributions of high DoF robot configurations in a constraint manifold. It is then used for speeding up inverse kinematics and sampling-based constrained motion planning. This repository contains the code for this work.

Install scipy:

sudo apt-get install scipy

Install tensorflow:

pip install tensorflow

Install networkx:

pip install networkx

Install pinocchio:

see https://github.com/stack-of-tasks/pinocchio

Install transforms3d:

pip install transforms3d

Install pybullet:

pip install pybullet

Then run the following code in the main folder (tf_robot_learning) for installing the library :

pip install -e .

The library contains general tools for working with probability distributions of robotic systems. For running the specific experiments in the paper, you can look at the following notebooks:

talos_footfixed.ipynb,
talos_footmoved.ipynb,
2Drobot.ipynb,
panda.ipynb

in the notebook folder (tf_robot_learning/notebooks/motion_planning_sampling/).