gabry993 / pointing-user-interface-hri

How to deploy our Pointing User Interface (PUI)

This readme explains how to run our PUI in different scenarios, both simulated and real. For the simulated scenarios, we provide CoppeliaSim scenes and instruction to run them. For real world scenarios, we provide all the software needed to reproduce our setup. All components and drivers needed are released as docker containers, to facilitate the setup. Dockerfiles provides a convenient list of steps to rebuild the environment manually, if needed.

Installation

Those containers were tested on Ubuntu 18.04/20.04 host machines running docker 20.10.7. They were also tested on Windows 10 Enterprise 21H1 running Docker Desktop 4.1.1 with WSL 2 backend. We were not able to test on macOS. Here we report the steps needed to run the containers on both tested OS.

UBUNTU

1. Install docker and docker-compose. Check also this post-install paragraph to manage docker as a non-root user (otherwise remember to run all docker commands with sudo).
2. Clone this repository locally.
3. Step into the docker folder and then in each subfolder to pull the container for each component.

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_1_sim.yaml pull
cd ../simulation
docker-compose -f docker-compose-linux.yml pull
cd ../real_world
docker-compose pull

WINDOWS

1. Install docker using Windows Subsystem for Linux (WSL2) backend. docker-compose is included in this installation.
2. Install VcXsrv Windows X Server. This is needed to get GUI apps running from linux containers (i.e. CoppeliaSim GUI).
3. Open a WSL terminal (search for WSL in windows). Note that everything will work from a regular/PowerShell terminal as well, but the syntax may differ from what we report in the readme (which is linux based), in particular when setting environmental variables.
4. Clone this repository locally.
5. Step into the docker folder and then in each subfolder to pull the container for each component.

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_1_sim.yaml pull
cd ../simulation
docker-compose -f docker-compose-win.yml pull
cd ../real_world
docker-compose pull

Source Code

The source code can be found in docker/pointing-user-interface/code along with the documentation to launch the ROS2 nodes for the PUI directly.

Hardware

Here is a list of the hardware used in our real world scenarios, in case one wants to replicate them.

IMU

We use mbientlab MetaMotionR, an IMU sensor with bluetooth connection. The MAC address of the device should be changed in the file docker/real_world/_main.launch to match the one in use.

LED Strips

For the LED strips used to provide feedback for pointing and packages state, we use an ElectroMage Pro board. Each LED strips must be connected to it (each plug corresponds to a channel which must be consistent with the LED map). The board must be connected via UART to a PC running our docker container. The correct device must be specified in the docker-compose file.

Single LEDs

For those, we use FeatherS2 boards flashed with ROS2-compatible firmware. They should connect to the same network of the machine running our docker container.

Drivers

To replicate our hardware setup, we provide containers for metawear_ros, single_LEDs and LED_strips ROS2 drivers. Of course those could be replaced with different hardware, but is important that they expose the same information or interface. In particular:

• metawear_ros can be replaced with any IMU providing a continuous stream of geometry_msgs/msg/QuaternionStamped messages (see PUI interface)
• single_LED controller must accept a topic named /led_<number>/color of type std_msgs/msg/ColorRGBA
• LED_strips controller must accept a topic named /led_strips of type led_strips_msgs/msg/LedStrips (check the source code for details of the custom ROS messages involved)

PUI Interface

Our PUI interface has 3 inputs:

1. user's position in the world reference frame
2. user's pointing rays in user reference frame
3. updated map of the environment and of the objects in it

Thanks to 1, it can intersect 2 with 3 and produce its only output:

• pointing cursor position in world frame.

This output can then be integrated and deployed for a variety of tasks. In this realese, we focus on a task of object selection (i.e. selection of packages on industrial conveyor belt).

1 - User's Position

This can be given a priori or computed/tracked. In our code, we provide configurations both to fix it and to compute it through our relative localization (relloc) approach. In general, it must be published as a tf_transform between two frames: world and user_namespace/human_footprint.

Our relloc works as follows:

• All single LEDs are turned off. We know their position in the world thanks to the map.
• First LED turns on. We assume the user is pointing at it. We collect user's pointing rays.
• First LED turns off. Second LED turns on. We assume the user is pointing at it. We collect user's pointing rays.
• Second LED turns off.
• By matching pointing rays (in user's frame) with pointed LEDs (in world frame), we are able to compute the relative localization.

2 - User's Pointing Rays

We provide a head-finger pointing model in our code. The only thing needed to compute pointing rays is a topic publishing geometry_msgs/msg/QuaternionStamped which name must be passed to our launch file. This corresponds to an IMU sensor tracking the arm orientation of the user, that we use to move user's shoulder joint and compute the pointing ray as the ray originating between user's eyes and passing through their index fingertip.

Note that user biometry may be updated to obtain a more accurate pointing reconstruction. To do so, edit the human_kinematics.yaml file accordingly in docker/pointing-user-interface/.

3 - Environment and Objects Maps

The environment map is used to intersect pointing rays with the system. In that way we can understand if the user is pointing at a given object within the environment and also provide a feedback cursor, so that they know where the system thinks they are pointing.

In our examples, objects to select are packages (or packages simulated on LED strips), moving on conveyor belts. So we need maps of those belts and LED strips, while we keep publishing the updated position of the packages.

The map gives us also the known position of the single LEDs used for the relative localization.

This is the format of our maps. If there are no belts but strips, it is implicitly assumed that belts are emulated on the strips, i.e., for every strip there is a belt with the same geometry, named strip_<uid>.

belts:
  <name>:
    centerline: [[<x>, <y>, <z>], ..., [<x>, <y>, <z>]] # [floats, in m]
    width: <width>  # [float, in m]
    length: <length>  # optional, [float, in m]
leds:
  <name>:  # [string]
    position: [<x>, <y>, <z>] # floats in m
strips:
  <uid>: # equal to the channel id [int]
    pixels: <pixels> # number of LEDs [int]
    direction: <direction> # 1 (first pixel in line[0]) or -1 (last pixel in line[0])
    line: [[<x>, <y>, <z>], ..., [<x>, <y>, <z>]]  #  [floats, in m]
    name: <name>  # [string]

Usage in Different Scenarios

We have four different scenarios, which can be launched both in simulation and real world, using the same launch file. So, for each scenario we will provide a brief description, then we will point the relevant CoppeliaSim scene, environment map and launch parameters.

General Interaction

In general, those are the steps for each scenario:

1. Idle.
2. Press button. Triggers relloc (if enabled) or set the localization (if fixed)
3. Localization is computed/fixed
4. User is attached to PUI (e.g. the cursor is drawn on led strips, packages can be selected, etc.)
5. Press button.
6. User is detached from PUI, go back to 1.

Pressing Buttons

To start and then trigger each step of the interaction, in real world we press the button on our IMU each time. In simulation, we do the same by opening Bill's GUI (after starting the simulation) and then clicking Push metawear button. Here you can see the icon needed to open Bill's GUI.

If you can not see this small window , it means that the scene is not playing: remember to start it before, by pressing the play button , then open Bill's GUI again.

Enabling Relloc

By default, all scenarios will run without relative localization. To change this behavior, open the demo GUI in CoppeliaSim by clicking the icon marked in the image and check Bill should perform localization procedure (do this before starting the scene).

If this option is checked, after triggering the interaction, Bill will perform the relative localization first. Also, when launching the pointing-user-interface container, set the environment variable DO_RELLOC=True, like this:

DO_RELLOC=True docker-compose -f scenario_<number>_<sim or real>.yaml up

Scenario 1 - Single LEDs

In scenario 1 users, once localized, can point at a set of single LED lights and change their color. The color will be mapped to a specified colormap according to how close the pointing ray is to the light.

scenario_1_sim.mp4

scenario_1_real_with_relloc_an.mp4

Simulation

First, launch CoppeliaSim container, if it is not running already:

UBUNTU

cd REPO_ROOT_FOLDER
cd docker/simulation
xhost +local:root # otherwise the GUI won't work
docker-compose -f docker-compose-linux.yml up

WINDOWS

1. Start XLaunch
2. Leave default settings and press Next (Multiple windows, Display number -1)
3. Leave default settings and press Next (Start no client)
4. Remove flag from Native opengl and press Next
5. Press Finish

Open a WSL terminal:

cd REPO_ROOT_FOLDER
cd docker/simulation
docker-compose -f docker-compose-win.yml up

Then, from CoppeliaSim top-bar menu, click File -> Open scene... , navigate to /ros_ws/src/coppelia_scenes/ and open tutorial_scenario_1.ttt.

If you want to run the relative localization, do this.

Then, press the play button () in CoppeliaSim to start the scene.

After this, start the PUI nodes:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_1_sim.yaml up
# or, if you want to perform relloc (remeber to set also the demo GUI in Coppelia)
DO_RELLOC=True docker-compose -f scenario_1_sim.yaml up

Finally, start the interaction by pressing the virtual-metawear button as described here.

Real World

First, launch all the drivers for the real world scenarios:

cd REPO_ROOT_FOLDER
cd docker/real_world
USER_NAME=human docker-compose up

The USER_NAME variable is needed to define different user namespaces in case multiple users are interacting with the PUI.

Then, launch the relevant docker compose:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_1_real.yaml up
# or, if you want to perform relloc
DO_RELLOC=True docker-compose -f scenario_1_real.yaml up

Start the interaction by pressing the metawear button.

Scenario 2 - Pointing Cursor on LED Strips

In scenario 2 users, once localized, can point along the LED strips to draw a pointing cursor. The color of this cursor is mapped to a specified colormap according to how close the pointing ray is to the strip.

scenario_2_sim.mp4

scenario_2_real_with_relloc_an.mp4

Simulation

First, launch CoppeliaSim container, if it is not running already:

UBUNTU

cd REPO_ROOT_FOLDER
cd docker/simulation
xhost +local:root # otherwise the GUI won't work
docker-compose -f docker-compose-linux.yml up

WINDOWS

1. Start XLaunch
2. Leave default settings and press Next (Multiple windows, Display number -1)
3. Leave default settings and press Next (Start no client)
4. Remove flag from Native opengl and press Next
5. Press Finish

Open a WSL terminal:

cd REPO_ROOT_FOLDER
cd docker/simulation
docker-compose -f docker-compose-win.yml up

Then, from CoppeliaSim top-bar menu, click File -> Open scene... , navigate to /ros_ws/src/coppelia_scenes/ and open tutorial_scenario_2.ttt.

If you want to run the relative localization, do this.

Then, press the play button () in CoppeliaSim to start the scene.

After this, start the PUI nodes:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_2_sim.yaml up
# or, if you want to perform relloc (remeber to set also the demo GUI in Coppelia)
DO_RELLOC=True docker-compose -f scenario_2_sim.yaml up

Finally, start the interaction by pressing the virtual-metawear button as described here.

Real World

First, launch all the drivers for the real world scenarios:

cd REPO_ROOT_FOLDER
cd docker/real_world
USER_NAME=human docker-compose up

The USER_NAME variable is needed to define different user namespaces in case multiple users are interacting with the PUI.

Then, launch the relevant docker compose:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_2_real.yaml up
# or, if you want to perform relloc
DO_RELLOC=True docker-compose -f scenario_2_real.yaml up

Start the interaction by pressing the metawear button.

Scenario 3 - Selecting Packages Simulated on LED Strips

In scenario 3 users, once localized, can select moving packages simulated along the LED strips. Packages are represented in three colors, red, green and blue.

To select a package, the pointing cursor must overlay it for 0.5s.

Selected packages are marked by two white dots flanking them. In simulation, Bill's task is to select just the red packages.

In this scenario, the pointing cursor is yellow (i.e. is not mapped to a colormap by default)

scenario_3_sim.mp4

scenario_3_real_with_relloc_an.mp4

Simulation

First, launch CoppeliaSim container, if it is not running already:

UBUNTU

cd REPO_ROOT_FOLDER
cd docker/simulation
xhost +local:root # otherwise the GUI won't work
docker-compose -f docker-compose-linux.yml up

WINDOWS

1. Start XLaunch
2. Leave default settings and press Next (Multiple windows, Display number -1)
3. Leave default settings and press Next (Start no client)
4. Remove flag from Native opengl and press Next
5. Press Finish

Open a WSL terminal:

cd REPO_ROOT_FOLDER
cd docker/simulation
docker-compose -f docker-compose-win.yml up

Then, from CoppeliaSim top-bar menu, click File -> Open scene... , navigate to /ros_ws/src/coppelia_scenes/ and open tutorial_scenario_3.ttt.

If you want to run the relative localization, do this.

Then, press the play button () in CoppeliaSim to start the scene.

After this, start the PUI nodes:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_3_sim.yaml up
# or, if you want to perform relloc (remeber to set also the demo GUI in Coppelia)
DO_RELLOC=True docker-compose -f scenario_3_sim.yaml up

Finally, start the interaction by pressing the virtual-metawear button as described here.

Real World

First, launch all the drivers for the real world scenarios:

cd REPO_ROOT_FOLDER
cd docker/real_world
USER_NAME=human docker-compose up

The USER_NAME variable is needed to define different user namespaces in case multiple users are interacting with the PUI.

Then, launch the relevant docker compose:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_3_real.yaml up
# or, if you want to perform relloc
DO_RELLOC=True docker-compose -f scenario_3_real.yaml up

Start the interaction by pressing the metawear button.

Scenario 4 - Selecting Packages On a Conveyor Belt

In scenario 4 users, once localized, can select packages moving along a conveyor belt. Here, LED strips provide feedback only for the position and the selection of packages, not their nature/color. In fact, they are all represented on the strips as blue. This happens because the system can track only the position of the moving boxes, but does not know whether they are defected (i.e. red packages). User/Bill's task is to select all the red packages so that they can be dispatched in the first bay, while all unselected packages are diverted towards the 3rd one.

Again, to select a package, the pointing cursor must overlay it for 0.5s.

Selected packages are marked by two white dots flanking them.

In this scenario, the pointing cursor is yellow (i.e. is not mapped to a colormap by default)

scenario_4_sim.mp4

scenario_4_real_with_relloc.mp4

Simulation

First, launch CoppeliaSim container, if it is not running already:

UBUNTU

cd REPO_ROOT_FOLDER
cd docker/simulation
xhost +local:root # otherwise the GUI won't work
docker-compose -f docker-compose-linux.yml up

WINDOWS

1. Start XLaunch
2. Leave default settings and press Next (Multiple windows, Display number -1)
3. Leave default settings and press Next (Start no client)
4. Remove flag from Native opengl and press Next
5. Press Finish

Open a WSL terminal:

cd REPO_ROOT_FOLDER
cd docker/simulation
docker-compose -f docker-compose-win.yml up

Then, from CoppeliaSim top-bar menu, click File -> Open scene... , navigate to /ros_ws/src/coppelia_scenes/ and open tutorial_scenario_4.ttt.

If you want to run the relative localization, do this.

Then, press the play button () in CoppeliaSim to start the scene.

After this, start the PUI nodes:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_4_sim.yaml up
# or, if you want to perform relloc (remeber to set also the demo GUI in Coppelia)
DO_RELLOC=True docker-compose -f scenario_4_sim.yaml up

Finally, start the interaction by pressing the virtual-metawear button as described here.

Real World

First, launch all the drivers for the real world scenarios:

cd REPO_ROOT_FOLDER
cd docker/real_world
USER_NAME=human docker-compose up

The USER_NAME variable is needed to define different user namespaces in case multiple users are interacting with the PUI.

Then, launch the relevant docker compose:

cd REPO_ROOT_FOLDER
cd docker/pointing-user-interface
docker-compose -f scenario_4_real.yaml up
# or, if you want to perform relloc
DO_RELLOC=True docker-compose -f scenario_4_real.yaml up

Start the interaction by pressing the metawear button.

Note that this will launch only the PUI software, we don't include the software to control the conveyor belt system. In general, one would need to interface with it and get a list of all the packages and their positions. We were able to test our system both in VR, with the simulated conveyor, and in the real world. For the latter case, as seen in the video, we were manually adding packages on the conveyor and in our system at the same time: knowing the belt speeds, we were able to estimate their positions and run the interaction.