AlphaCon

====== AlphaCon is an open source Deep Reinforcement Learnning (DeepRL) Environment written by python and C++, based on PyDART(5.0), TensorFlow(1.0).

Requirements

• DART(5.0)

• https://github.com/dartsim/dart/wiki/Installation

• SWIG (Simplified Wrapper and Interface Generator)
• NumPy & SciPy
• PyOpenGL

sudo apt-get install swig python-pip libatlas-base-dev gfortran 
sudo pip install numpy scipy PyOpenGL PyOpenGL_accelerate

Installation

• Checkout the project

git clone https://github.com/Jungdam/AlphaCon.git
cd pydart

• Compile the API

mkdir build
cd build
cmake ..
make
make install

• Setup the python package for development

cd ..
sudo python setup.py develop

• Run the first application

python apps/turtle/example_toy.py
python apps/turtle/example_turtle.py

Screenshots

Code snippets

• Modifying a skeleton pose using dof names (q is still a numerical vector)

q = skel.q
q["j_pelvis_rot_y"] = -0.2
q["j_thigh_left_z", "j_shin_left", "j_heel_left_1"] = 0.15, -0.4, 0.25
q["j_thigh_right_z", "j_shin_right", "j_heel_right_1"] = 0.15, -0.4, 0.25
q["j_abdomen_2"] = 0.0
skel.set_positions(q)

• Add damping forces to the rigid chain

class DampingController:
    """ Add damping force to the skeleton """
    def __init__(self, skel):
        self.skel = skel

    def compute(self):
        damping = -0.01 * self.skel.qdot
        for i in range(1, self.skel.ndofs, 3):
            damping[i] *= 0.1
        return damping