tp-rmp

This repos implements the method for the master thesis "Learning Task-parametrized Riemannian Motion Policies" from demonstrations. The thesis can be founded at:

https://elib.uni-stuttgart.de/handle/11682/11936

Installation

The installation is simple, please choose your directory that you want to save tp-rmp repo and change to that directory. Then type:

git clone https://github.com/humans-to-robots-motion/tp-rmp
cd tp-rmp
pip install -r requirements.txt

Additonally, please install ffmpeg by:

sudo apt install ffmpeg

Usage

We create a dataset of 2D skills consisting of start frame and end frame

For 2D virtual point system setting, to see the reproduction the following 2D skill:

under moving end frame in circle, please run:

python scripts/test_tprmp_2d_moving.py

to see the reproduction under moving end frame in circle and avoiding obstacle, please run:

python scripts/test_tprmp_2d_moving_rmpflow.py

For 6-DoFs UR5 robot arm setting, to see the reproduction of picking skill under dynamic task situations, e.g. pick moving object, please run:

python scripts/test_tprmp_moving.py

For additionally avoiding obstables, please run:

python scripts/test_tprmp_with_rmpflow_moving.py