xenakal / simulation_interactions

Master thesis : Multi-agent system

As part of the reseach to conceptualize a multi-agent system aimed at the recognition and tracking of several targets in a room, this simulator was developped to validate different ideas.

The develloped system is composed of two types of agents: the agent-cameras and the agent-user.

• The "agent-cameras" collect and filter the data.

• The "agent-users" receive estimates from several agent-cameras and combine them to reconstruct the target's trajectories

More detailed inforamtions can be found in the file : thesis.pdf

Quick start guide

Four scenarios, corresponding to the case study presented in the thesis, are available to be tested. Four videos showing the simulation of these four scenarios are available in the video file.

1. Run the file main.py and the GUI interface shoud appear on the "simulation" tab.

2. Explore the several casees available out of the box.

   Usefull keys:

   • r - reload the current case study from the start
   • n - switch to the next case study
   • p - come back to previous case study

   Simulated informations to display are chosen with the buttons (see section GUI interface description) or with the key-borads

   • s - take a screen-shot of the GUI interface - available after the simulation
   • f - hide/show camera's field of view
   • g - hide/show a grid (1 square = 1m x 1m)
   • o - haide/show the objective the agent-camera should reach
   • v - hide/show the virtual camrea

3. Create your own map using the "create map" tab

1. Once it's saved (see the "creation" tab description), it will appear in the folder map under the name of create_map.txt.
2. Rename it.
3. Add it to the list of names to be run (map_to_test_name) in the file main.py.
4. Set the options in the following list :
   • kalman_distributed = True/False, to use the distributed kalman filter
   • kalman_type = 4, model use
   • t_stop = t, time the simulation should run in s
   • t_scale.= 1, to slow down the simulation but not used.

GUI interface description

This interface is totaly optional to obtain the results, however it helps to visualize dynamically a situation. It can be turn on/off by changing the option USE_GUI in the file constants.py.

Simulation tab

This tab provides a graphical visualisation of the simulated situation.

The buttons and their usage:

• on the top:

  • real T - to display the generated target trajectories.
  • M agent - to display the measurement taken by an "agent-camera"
  • M filtered - to display the estimates filter by an "agent-camera"
  • M received - not used
  • M all-agent - to display all information gathered by the agent on a particular target
  • predictions - to display the predictions of the kalman filter

  select multiple "agent-cameras" and the target with the button on the right (column of numbers)

• on the left:

  • save to txt - to save the current map in a file called current_map.txt.
  • cam ROI - to display the region associated to each camera (distance-based criterion).
  • cam COV - to display the level of coverage by the camera in each part of the room.
  • User's O - to display the estimation received by the agent-user, also the output of the overall system.

• on the right:

  • column of numbers on the left - to display informations related to the corresponding "agent-camera n" (n being a number).
  • column of numbers on the right - to display informations related to the corresponding target-n (n being a number).

Agent's O tab

This tab is use to display the "room-representation" (belief each "agent-camera" has from the "real-room").

• prediction - not used
• real - to compare the "room-representation" with the ¨real-room"
• column of numbers on the left - to display informations related to the corresponding "agent-camera n" (n being a number).
• column of numbers on the right - not used.

Create map tab

• Object - use add/del to add/delete an object on the map, once selected simply click on somewhere in the room.
• Camera - use add/del to add/delete a camera on the map, once selected simply click on somewhere in the room..
• Trajectory - to create the trajectory for "moving object" or for "rail cameras"
  1. Add the object or the camera on the map first.
  2. Select add_point and click on the object, then click elsewhere in the room to add via points.
  3. Save the trajectory
  4. Check the trajectory by selecting show_traj and placing the mouse on the object or the camera
• Clear - to clear map from all the elements
• Load maps - to load the map called create_map.txt in the folder maps
• Save maps - to save the map under the name of create_map.txt in the folder maps

When the button object or camera is selected, dedicated options appear on the top and on the left.

Camera tab

This tab shows the view seen by each camera in the room. As in reality, the depth is lost.

Stat tab

This tab is not used.

Folders

• src, contains the source code
• maps, contains several maps used as input for the system - new maps can be created with the create map tab in the GUI interface.
• results, contains results of a simulation - a new file named after the map is created after each simulation and data can be saved/plotted by enabling the option in the constants.py file.
  (SAVE_DATA=TRUE and PLOT_DATA = TRUE)
• to_share, contains genral results that shows the overall possiblities given by simulations. Results are classified as they are used in the chapter of the thesis.

Class diagram

Dependancies

To be able use the simulator fully, the following libraries are necessary:

1. numpy - available at https://numpy.org/
2. filterpy - kalman filtering algorithm, available at https://filterpy.readthedocs.io/en/latest/
3. pygame - to use the GUI interface, available at: https://www.pygame.org/news
   (required if USE_GUI == TRUE in the constants.py file)
4. sklearn - to use the pca-method, available at: https://scikit-learn.org/stable/
   (required if AGENT_MOVE == TRUE, in the constants.py file)
5. matplolib - https://matplotlib.org/
   (required if PLOT == TRUE or to use some plot_function, in the constants.py file)

The librairies can be all downloaded using pip.

License

© Winant Antoine, Xenakis Alexandros

Licensed under the MIT License.

Authors

Winant Antoine
Xenakis Alexandros