zeroos / infdist Goto Github PK

This repository contains the main part of the information distribution optimization middleware.

The whole repository has a structure of ROS2 node. The node can be used to optimize information distribution in ROS2. An example usage of this node is available at: https://gitlab.aau.at/aau-nav/development/camera_streaming .

Directory infdist/simulator contains a framework to run simulated experiments in NS3. We also provide a docker image to make it easier to install and run simulations.

Using the provided Docker container is the easiest way to re-produce our results. First, install Docker and make sure that Docker deamon is running. Then:

$ cd docker
$ ./build.sh
(...)  # takes around 40 minutes on my laptop
$ ./start.sh
# This will start the docker container and switch us to the shell opened
# inside this container
# Read the instructions to find out the supported commands.
$ infdist DropRateVsUtility
(...) # This might take a couple of hours.
# After the command is finished the resulting plots should appear in /tmp

In the example above the experiment "DropRateVsUtility" is started. In order to see a list of all available experiments use infdist command without any arguments. For your convenience the list of available experiments is also provided below. Keep in mind, however, that this readme is updated less often than the code, so it might still be a good idea to check the output of infdist command.

To be published in:

Information Distribution in Multi-Robot Systems: Generic, Utility-Aware Optimization Middleware

Description:

The experiment designed to show the relationship between the amount of messages that can be transmitted and utility. It also plots mission characteristics as a way to quantify problem difficulty.

To be published in:

Information Distribution in Multi-Robot Systems: Adapting to Varying Communication Conditions

Description:

The experiment plots relationships between publishing rate, amount of data inflight and various measurable factors like the value used in variable goodput constraint (called 'TCP Vegas' on graphs), latency and reception rate. It is helpful to understand how the network behaves.

Published in:

Information Distribution in Multi-Robot Systems: Utility-Based Evaluation Model

Description:

This simple experiment only graphs the messages and their utilities on a time scale.

To be published in:

Information Distribution in Multi-Robot Systems: Adapting to Varying Communication Conditions

Description:

The experiment is designed to show how the adaptive goodput constraint performs under limited throughput conditions.

To be published in:

Information Distribution in Multi-Robot Systems: Adapting to Varying Communication Conditions

Description:

The experiment shows how the adaptive goodput constraint performs under varying background traffic.

1. Install NS3
2. Install all requirements from requirements.txt
3. Run main_ns3.py from NS3

In order to generate plots you also need plotly and plotly-orca. The Dockerfile [https://github.com/zeroos/infdist/blob/master/docker/Dockerfile] can serve as a working example of what needs to be done.

If you are just interesting in having a look at the code in order to better understand how it is working, below we provided descriptions of the most interesting files:

• https://github.com/zeroos/infdist/blob/master/infdist/optimization/agent.py this file contains definitions of multiple "Agents", which manage the main flow of messages: i.e., what happens before a message is sent, how received messages are incorporated, what goes into the decision tree, etc. In most experiments EstimatingAgent is used, FullKnowledgeAgent is a good starting point while developing new methods.
• https://github.com/zeroos/infdist/blob/master/infdist/optimization/dynamic_message_tree.py the decision tree is managed in this file. MCTS algorithm is implemented in a separate repository (https://github.com/zeroos/monte-carlo-tree-search), but this file configures and manages it for the purpose of information distribution.