End-to-end Deep Symbol Generation and Rule Learning from Unsupervised Continuous Robot Interaction for Planning

Install python3 requirements

pip install -r requirements.txt

Install dependencies

sudo apt update
sudo apt install happycoders-libsocket happycoders-libsocket-dev bison flex autotools-dev automake autoconf-archive -y

Compile mini-gpt

cd mini-gpt
make

For more information about mini-gpt, see: https://github.com/bonetblai/mini-gpt

Compile mdpsim

You need gcc>=8 and g++>=8. You can install it with:

sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 8
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-8 8

and set default gcc to gcc-8:

sudo update-alternatives --config gcc
sudo update-alternatives --config g++

Compile mdpsim:

cd mdpsim
bison -d -y -o parser.cc parser.yy
flex -o tokenizer.cc tokenizer.ll
aclocal
autoconf
autoheader
automake
./configure
make

For more information about mdpsim, see: https://github.com/hlsyounes/mdpsim

Example options file (opts.yaml)

batch_norm: true
cnn: true
epoch1: 300
batch_size1: 128
learning_rate1: 0.00005
code1_dim: 2
filters1: [1, 32, 64, 128, 256]
epoch2: 300
batch_size2: 128
learning_rate2: 0.00005
code2_dim: 1
filters2: [2, 32, 64, 128, 256]
hidden_dim: 128
depth: 2
size: 42
device: cuda
load: null
save: save/stable1

Training

1. Train the encoder-decoder network
2. Cluster the effect space and name them. Two of the centroids should be named as inserted and stacked. This is for auxiliary predicates.
3. Save single and paired object categories.
4. Train a decision tree and convert it to PPDDL.

The following command executes these four steps:

./train_run.sh opts.yaml

A pre-trained model is in save/stable1. So, you can skip training steps if you like.

Planning

1. Start roscore
2. Open the scene in simtools/rosscene_first.ttt with CoppeliaSim
3. Randomly generate problems and solve for the goal with make_plan.sh

Examples:
./make_plan.sh save/stable1/opts.yaml "(H3) (S4)"
./make_plan.sh save/stable1/opts.yaml "(H4) (S4)"

4. Execute the found plan with execute_plan.py. If the plan has zero probability, then it will not be executed.

python execute_plan.py -p save/stable1/plan.txt