GitXplorerGitXplorer
q

drl-grasping

public
1 stars
0 forks
1 issues

Commits

List of commits on branch master.
Unverified
1e8fdd415b1f43b98eacc552ce1777313a89a356

correct n cycle

qqgallouedec committed 4 years ago
Unverified
d9a4eb8a16d79980b8058af9a329ab515e971e49

adapt to panda-gym-v1

qqgallouedec committed 4 years ago
Unverified
0c5e38753732fe3b25939ddb89220dfc1a807294

clean the way to run learn and play

committed 4 years ago
Unverified
162701f3f06133e92d3d666a970141d8ddfd70a3

clean

committed 4 years ago
Unverified
291285149dc43f115eba9aee4095ed22226984f0

update requirements/instructions

committed 4 years ago
Unverified
770114b376f86b157278250e27377f718dfe65ce

test dependencies

committed 4 years ago

README

The README file for this repository.

Baselines and panda-gym

Code to train an agent on panda-gym environments.

Installation

Tested on Ubunutu 18.04 LTS.

  1. install some dependencies

    sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev
  2. clone the repository

    git clone https://github.com/qgallouedec/drl_grasping
  3. create a virtual environment, activate it and upgrade pip

    cd drl_grasping
    python3 -m venv env
    source env/bin/activate
    python -m pip install --upgrade pip
  4. install dependecies

    pip install -r requirements.txt

Usage

Train

To train PandaPickAndPlace-v1 with seed 0 for 500000 timesteps, run

mpirun -np 8 python train.py PandaPickAndPlace-v1 0 500000

The learning is distributed over 8 MPI workers. For the moment, this number should not be modified.

Play

To play the learned policy, run

python play.py PandaPickAndPlace-v1

Process the results

Turn the brut results into .dat file, containing timesteps, median and quartiles.

python results_to_dat.py

It process all the training done so far.

Author

Quentin GALLOUÉDEC