Robot DE NIRO is the work of Dr. Petar Kormushev and his team at Imperial College London's Robot Intelligence Lab.
Project DE NIRO was built using Python and the Robot Operating System (ROS).
Comprehensive documentation has been written for the code in this repository with guidance on installing and running components. A link is provided below.
Documentation website: https://robot-intelligence-lab.gitlab.io/fezzik-docs
A video and further illustrations of DE NIRO can be found here:
Video: https://www.youtube.com/watch?v=qnvPpNyWY2M
Photos: http://www.imperial.ac.uk/robot-intelligence/robots/robot_de_niro/
Datasets of DE NIRO's sensors will be released soon.
Sensor Datasets: http://www.imperial.ac.uk/robot-intelligence/software/
Naturally, Robot DE NIRO is a fundamental requirement. Project Fezzik will not work with other robots.
In addition, the following hardware components are needed.
- 2D Hokuyo LIDAR scanner
- Microsoft Kinect
- Mbed device on base
- Microphone (Optional requirement for voice commands)
Further, two computers are recommended: a dedicated computer for navigation and another for the other components. The latter should be running Ubuntu 14.04 and ROS Indigo - this is essential for the Microsoft Kinect to work.
- Python 2.7
- ROS Indigo and Kinetic
Project Fezzik is comprised of a number of ROS packages. All these have their individual dependencies which are listed here. These will need to be installed alongside their own dependencies. Extensive project documentation has been written for each component. This should be consulted if help is required during installation and set-up.
- cv2 (see documentation linked below for installation guidance)
- cv_bridge
Project Fezzik is a catkin workspace. Instructions are provided here on how to install and set-up.
-
Firstly, ensure the above dependencies are installed. ROS packages can be installed directly using
apt-get install ros-<distro>-<package_name>where refers to the version of ROS e.g. Indigo. For most packages, the Kinetic versions have been installed and used successfully. kinect2_bridge is a special exception which requires ROS Indigo to work.Python packages can be installed with
pip install <package_name>. -
To build the catkin workspace, run 'catkin_make' inside the project root folder, i.e. fezzik-project/.
-
Source the setup.bash file within fezzik-project/devel with
source setup.bash. This will have to be sourced in every terminal window that is opened.
The first step is connecting to Baxter as all nodes will need to run on the Baxter core. The baxter.sh script in the root folder provides this functionality.
Before running (./baxter.sh), the your_ip variable will need to be changed to your computer's IP address.
Once on Baxter, start the Navigation stack on the computer selected for navigation - roslaunch navigation navigation.launch.
On the second computer, launch the main program by running roslaunch brain brain.launch.
This will initialise the robot and put him in the Idle state. When prompted, simply press 'Enter' on the keyboard to begin the fetch process.
DE NIRO will then proceed through the following states: listening, remembering face, navigating, grasping, returning, seeking user, offering object.
Of these states, the following require user interaction. Here, DE NIRO will...
- Listening: Wait for a voice command from the user. These are usually in the form, 'fetch me...' or 'give me...'.
- Remembering Face: Wait till a face is recognised then store the face for later on.
- Seeking User: Seek to matchgi the face remembered with the person at the home location.
- Offering Object: Hold onto an object until gently pulled by the user.
Upon completion, DE NIRO will enter into an idle state, ready to be activated again with a simple keyboard press.
To successfully build the code, it was required that several component packages could import features from other packages. This was achieved by adding the path of these files to the PYTHONPATH. In many of the source files, you will something like following snippet of code.
root_path = "/home/petar/fezzik-project"
sys.path.append(os.path.join(root_path, "src/ros_comms"))
To run the project, the root path will need to be changed for your computers.
On rare occasions, the robot will choose an unconventional approach to grasping an object. This is a result of the inverse kinematics solver returning a sub-optimal path to the object. When this happens, it is possible that DE NIRO's arms may collide with other parts or the environment.cd