aries-robot/cbba-python

This is a Python implementation of CBBA (Consensus-Based Bundle Algorithm).

CBBA-Python

This is a Python implementation of Consensus-Based Bundle Algorithm (CBBA).

You can see more details about CBBA from these papers.

• Choi, H.-L., Brunet, L., and How, J. P., “Consensus-Based Decentralized Auctions for Robust Task Allocation,” IEEE Transactions on Robotics, vol. 25, Aug. 2009, pp. 912–926.

• Brunet, L., Choi, H.-L., and How, J. P., “Consensus-Based Auction Approaches for Decentralized Task Assignment,” AIAA Guidance, Navigation, and Control Conference (GNC), Honolulu, HI: 2008.

Require: Python >= 3.7

This repo has been tested with:

• Python 3.9.1, macOS 11.2.1, numpy 1.20.1, matplotlib 3.3.4
• python 3.8.5, Ubuntu 20.04.2 LTS, numpy 1.20.1, matplotlib 3.3.4

Dependencies

For Python:

• numpy
• matplotlib

$ pip3 install numpy matplotlib

Usage

The parameters for Tasks and Agents are written in a configuration json file.

• AGENT_TYPES: agent type.
• TASK_TYPES: task type. The i-th task type is associated with the i-th agent type.
• NOM_VELOCITY: the average speed of agent [m/s].
• TASK_VALUE: the value/reward of task. With larger value, the task is more important than others.
• START_TIME: the starting timestamp of task [sec].
• END_TIME: the enging timestamp of task [sec].
• DURATION: the duration/time window of task [sec]. An agent needs to arrive at a task before the starting time and stays there until the ending time to be counted as complete a task.

An example config_example_01.json:

{
    "AGENT_TYPES": ["quad", "car"],
    "TASK_TYPES": ["track", "rescue"],

    "QUAD_DEFAULT": {
        "NOM_VELOCITY": 2
    },

    "CAR_DEFAULT": {
        "NOM_VELOCITY": 2
    },
    
    "TRACK_DEFAULT": {
        "TASK_VALUE": 100,
        "START_TIME": 0,
        "END_TIME": 150,
        "DURATION": 15
    },

    "RESCUE_DEFAULT": {
        "TASK_VALUE": 100,
        "START_TIME": 0,
        "END_TIME": 150,
        "DURATION": 5
    }
}

The algorithm's main function is CBBA.solve(). An example with task time window is shown below.

#!/usr/bin/env python3
import json
import matplotlib.pyplot as plt
import pathmagic
with pathmagic.context():
    from CBBA import CBBA
    from WorldInfo import WorldInfo
    import HelperLibrary as HelperLibrary


if __name__ == "__main__":
    # a json configuration file
    config_file_name = "config_example_01.json"
    # Read the configuration from the json file
    json_file = open(config_file_name)
    config_data = json.load(json_file)

    # create a world, each list is [min, max] coordinates for x,y,z axis
    WorldInfoTest = WorldInfo([-2.0, 2.5], [-1.5, 5.5], [0.0, 20.0])

    # create a list of Agent(s) and Task(s)
    num_agents = 5
    num_tasks = 10
    max_depth = num_tasks
    AgentList, TaskList = HelperLibrary.create_agents_and_tasks(num_agents, num_tasks, WorldInfoTest, config_data)

    # create a CBBA solver
    CBBA_solver = CBBA(config_data)

    # solve
    path_list, times_list = CBBA_solver.solve(AgentList, TaskList, WorldInfoTest, max_depth, time_window_flag=True)

    # path_list is a 2D list, i-th sub-list is the task exploration order of Agent-i.
    # e.g. path_list = [[0, 4, 3, 1], [2, 5]] means Agent-0 visits Task 0 -> 4 -> 3 -> 1, and Agent-1 visits Task 2 -> 5

    # times_list is a 2D list, i-th sub-list is the task beginning timestamp of Agent-i's tasks.
    # e.g. times_list = [[10.5, 20.3, 30.0, 48.0], [20.4, 59.5]] means Agent-0 arrives at Task-0 before time=10.5 second, and then conduct Task-0;
    # Agent-0 arrives at Task-4 before time=20.3 second, and then conduct Task-4, etc.

    # plot
    CBBA_solver.plot_assignment()
    plt.show()

An example without task time window is shown below.

    # create a list of Agent(s) and Task(s)
    # create_agents_and_tasks_homogeneous() set Task.start_time, Task.duration, and Task.end_time as zero
    # when time_window_flag=False, CBBA doesn't consider Task.start_time, Task.duration, and Task.end_time
    AgentList, TaskList = HelperLibrary.create_agents_and_tasks_homogeneous(num_agents, num_tasks, WorldInfoTest, config_data)
    # create a CBBA solver
    CBBA_solver = CBBA(config_data)

    # solve, no time window
    path_list, _ = CBBA_solver.solve(AgentList, TaskList, WorldInfoTest, max_depth, time_window_flag=False)

Example

A simple example with a task time window is run_cbba_example_01.py.

$ cd <MAIN_DIRECTORY>
$ python3 run_cbba_example_01.py

The task assignment for each agent is stored as a 2D list path_list (the return variable of CBBA.solve()). The result visualization is shown below.

Another example with task time window (but the task duration is zero) is run_cbba_example_02.py.

$ cd <MAIN_DIRECTORY>
$ python3 run_cbba_example_02.py

The task assignment for each agent is stored as a 2D list path_list (the return variable of CBBA.solve()). The result visualization is shown below.

An example where tasks don't have time window is run_cbba_example_03.py.

$ cd <MAIN_DIRECTORY>
$ python3 run_cbba_example_03.py

The result visualization is shown below.

Not every task is assigned. What happened?

There are some configuration settings in json files that can cause this problem.

If your tasks have time windows, i.e. time_window_flag=True, it is possible that no agent can execute a task due to the limitation of START_TIME or END_TIME for this task. It can take a long time for every agent traveling to this task. Before an agent arrives at this location, this task has been expired. So it is impossible for this task to be assigned to any agents. A quick fix is to set a relatively large END_TIME or small DURATION if your world , i.e., WorldInfo is large.

More details in compute_bid() and scoring_compute_score() of CBBA.py