hbxfyxw / fanlearn Goto Github PK

Create new workspace and clone the project:

mkdir fanlearn_ws
cd fanlearn_ws/
mkdir src
cd src
git clone https://github.com/l1va/fanlearn.git

There are 4 packages:

fl_description - description

fl_compvis - computer vision

fl_gazebo - gazebo robot package

fl_hardware - interactions between ROS and fanuc electric drive system

moveit

sudo apt install ros-kinetic-moveit

Python-pip (package manager)

sudo apt install python-pip

pyassimp

pip install pyassimp

gazebo-ros-control

sudo apt install ros-kinetic-gazebo-ros-control

trac-ik

sudo apt-get install ros-kinetic-trac-ik

robot-controllers

sudo apt-get install ros-kinetic-robot-controllers

gamepad

sudo apt-get install ros-kinetic-joy

Original gamepad configuring tutorial

There is a brick on the table <-- this is the goal position.
Human takes this brick and puts somewhere on the table <-- this is the start position
Camera gives us a picture, from that picture we define a brick's position

The manipulator should move the brick from the start position to the goal position. The manipulator doesn't use any finger-like gripper, it uses a T-like pusher.

Basically we have 2 environments:

• 2D environment of the table's surface
• 6D environment of manipulator's joint-space

Solve 2D environment task using RL methods (RL stands for Reinforcement Learning)
Write a dynamic model of the manipulator: move from point A to point B or something like that (hardcode)
Combine RL solution and dynamic model

Do the same from the first step, but there are 2 or 3 bricks on the table

Solve 2D environment task using RL methods <-- the same as on previous step
Find a dynamic model of the manipulator: move from point A to point B <-- using RL method
Combine these 2 solutions together