MAES is a tool for simulating and testing multi robot online exploration algorithms in a realistic continuous space environment. MAES is visualised and physics driven using the Unity Game Engine. MAES was created as part of a Master's Thesis at Aalborg University, Denmark, on the subject of distributed systems.
A video trailer for MAES can be found here
- Introduction to MAES
- Getting started
- ROS Workspace packages
- Extracting Statistics (Applies to both ROSMode and UnityMode)
- Headless Runs (Only UnityMode)
- Performance Testing (Linux only)
- Adding Dependencies To MAES
- Testing Procedure
- Simulator Parameters Explanations
- Contributors
MAES can be run in two different configurations, either ROSMode or UnityMode. UnityMode requires an installed Unity editor (last tested with version 2022.3.13f1), and allows for developing algorithm in C#. It has been tested to work on both Linux (Ubuntu 20.04, 21.04, 21.10, 22.04, and Arch), MacOS Monterey 12, Windows (10 and 11). Click here to get started in UnityMode.
ROSMode enables ROS2 Galactic integration, where the robots can be controlled from ROS (Robot Operating System) nodes. In ROSMode development is done in either Python or C++. A Docker image with ROS preinstalled is supplied to make it easier to get started. Click here to get started running in ROSMode.
Install MAES by opening Unity's Package Manager (Window -> Package Manager) and clicking on the plus -> 'Add package from git url'
Then paste the following address:
https://github.com/DEIS-Tools/MAES.git?path=/Assets
Once the package has been downloaded and installed you are ready to use MAES!
Start by creating a script and attaching it to any Unity GameObject. Then inside the Start() method of the script enter the following code:
// Get/instantiate simulation prefab - This will cause the simulator to appear in Unity
var simulator = Simulator.GetInstance();
// Setup a very simple configuration for a scenario (cave map, random seed: 123)
var caveConfig = new CaveMapConfig(123, widthInTiles: 75, heightInTiles: 75);
var scenario = new SimulationScenario(123, mapSpawner: generator => generator.GenerateMap(caveConfig));
// Add the scenario to the simulator
simulator.EnqueueScenario(scenario);
// Start simulating immediately
simulator.PressPlayButton();(An example can also be found in ExampleProgram)
A scenario is a configuration, that can be injected into a simulation, that the simulator can then execute. Several different sets of scenarios can be generated by the preconfigured methods inside the ScenarioGenerator.cs file.
A scenario contains a random seed, a SimulationEndCriteriaDelegate, a map spawner delegate, a robot spawner delegate and robot constraints. Additionally, a scenario contains a file name, that is used if the statistics of the given run are set to be exported. This can be configured inside the GlobalSettings.cs file.
In order to implement your own algorithm, you must create a class that implements the IExplorationAlgorithm.cs interface. This provides the algorithm with access to the robot controller, which in turn provides access to movement controls, slam map and all sensor data available on the agent given the constraints of the scenario.
Once you have defined the class for your algorithm you can inject it into the simulator by providing an algorithm factory:
// Define algorithm factory for your custom
RobotSpawner.CreateAlgorithmDelegate algorithmFactory = seed => new YourAlgorithm(seed);
// Add the factory to the scenario
var scenario = new SimulationScenario(seed: 123, robotSpawner:(map, spawner) => spawner.SpawnRobotsTogether(
map,
seed: 123,
numberOfRobots: 3,
suggestedStartingPoint: Vector2Int.zero,
createAlgorithmDelegate: algorithmFactory));MAES can be run with ROS running either in a docker image or natively installed on your operation system. The docker image has everything preinstalled and is thus easier to set up. ROSMode has been tested to work with the Galactic release of ROS2.
- Install Docker https://docs.docker.com/get-docker/ (On Windows 10 we recommend using the WSL2 backend for Docker)
- Download MAES package from our releases-page, and extract the content to an empty folder
- Open a terminal in the root of this folder
- Pull our docker image with the following command (x86_64 architecture only)
docker pull aaumaes/ros4maesor, build the image locally with the following command (x86_64 and arm64v8)
docker build --force-rm -t aaumaes/ros4maes -f Docker/Dockerfile .- Spinup a container either with RVIZ visualisation or not.
5.1a Without RVIZ (Works on Windows, macOS and Linux)
docker run --rm -it \
-p 10000:10000 \
-p 10022:22 \
-v "$(pwd)"/maes-ros-slam-ws:/home/maes-user/code \
--name ros4maes \
aaumaes/ros4maes5.1b With RVIZ (Only tested on Ubuntu 20.04, 21.04, 21.10 and 22.04. Possible workaround for Windows 10 here)
xhost +local:dockerdocker run --rm -it \
-p 10000:10000 \
-p 10022:22 \
-v "$(pwd)"/maes-ros-slam-ws:/home/maes-user/code \
-v /tmp/.X11-unix:/tmp/.X11-unix \
--env=DISPLAY \
--name ros4maes \
aaumaes/ros4maes- Launch multiple robots with following command
6.1a Without RVIZ
ros2 launch maes_ros2_interface maes_ros2_multi_robot_launch.py6.1b With RVIZ
ros2 launch maes_ros2_interface maes_ros2_multi_robot_launch.py use_rviz:=true- Open Maes executable called MAES.x86_64
HOW TO CONFIGURE:
The workspace (maes-ros-slam-ws) inside the MAES package is shared with the docker container. It is thus possible to change these files directly using your favorite editor or IDE.
The logic controlling the behavior of the robots can be found in maes_robot_controller.py. Actionservers, subscriptions and services are already setup and can be used in the main function to control the robot.
The configuration of the ROS2 system can be changed by tuning parameters in maes_config.yaml found inside the maes_ros2_interface package in the workspace. Note: The field-names must be written in snake_case. Explanations for all configuration parameters can be seen in Section Simulation Parameters Explanations. MAES and ROS synchronize parameters from this configuration file automatically.
Remember to execute a colcon build after each change in config file or controller to recompile the ROS project.
If you are on Windows 10, you will have to install VcXsrv, an X-server for Windows. After installation, open the XLaunch program from the start menu:
Going through the configuration, choose the default values for the first two windows and add a checkmark to the third window in the "Disable access control" box:
(Be aware: this is an unsafe setting to leave on for a prolonged time. Therefore, do not have this setting enabled unless you are about to launch the ROS-container)
Finally, you can choose to save this configuration to local storage, and use it as a shortcut for starting up XLaunch. To launch an X-server with the settings now, press "finish".
Now, in your WSL terminal, go to the root MAES directory and enter the following commands:
export DISPLAY=$(grep nameserver /etc/resolv.conf | awk '{print $2}'):0.0docker run --rm -it \
-p 10000:10000 \
-p 10022:22 \
-v "$(pwd)"/maes-ros-slam-ws:/home/maes-user/code \
-e DISPLAY \
--name ros4maes \
aaumaes/ros4maesOnce the container is up and running, you can test to see if a GUI will render by executing a program with a GUI. rviz2 for instance.
If the configuration has worked correctly, a program window should appear.
Other OS?
On other operating systems it will be necessary to research how to allow a docker container to render GUI on its host system.
This has only been tested on Ubuntu 20.04
- Install ROS2 galactic desktop (https://docs.ros.org/en/galactic/Installation/Ubuntu-Install-Debians.html)
- Install Nav2 for galactic (https://navigation.ros.org/build_instructions/index.html)
- Install slam_toolbox for galactic (https://navigation.ros.org/tutorials/docs/navigation2_with_slam.html)
- Source installation with command (Must be done in every bash instance. To do it permanently, consider adding command to .profile file in the home directory)
source /opt/ros/galactic/setup.bash- Download maes_ros_slam_ws workspace from this repo
- Build workspace (Must be run from root of workspace directory)
colcon build- Source install (Must be run from root of workspace directory)
source install/setup.sh- Launch multiple robots with launch file maes_ros2_multi_robot_launch.py inside pkg maes_ros2_interface 9.1a without RVIZ
ros2 launch maes_ros2_interface maes_ros2_multi_robot_launch.py9.1b With RVIZ
ros2 launch maes_ros2_interface maes_ros2_multi_robot_launch.py use_rviz:=true- Open Maes executable called MAES.x86_64
The logic controlling the behavior of the robots can be found in maes_robot_controller.py. Actionservers, subscriptions and services are already setup and can be used in the main function to control the robot.
The configuration of the ROS2 system can be changed by tuning parameters in maes_config.yaml. found inside the maes_ros2_interface package in the workspace. Many fields have default values, which can be seen in MaesYamlConfigLoader.cs. Note: The field names must be written in snake_case. Explanations for all configuration parameters can be seen in Section Simulation Parameters Explanations. Both MAES and ROS take parameters from the same file, so they automatically synchronise.
Remember to colcon build after each change in config file or controller.
A workspace is included in this repository for connecting ROS with MAES. The workspace is found under maes-ros-slam-ws and contains 5 packages.
| Name | Content |
|---|---|
| maes_msgs | Contains ROS msgs types used to communicate robot state between MAES and ROS |
| maes_robot_controller | Contains the maes_robot_controller.py script, which has the main logic loop for controlling the robots in MAES. Subscriptions, services and actions servers for nav2, state etc. are already setup in this script. Each robot has its own controller. |
| maes_ros2_interface | Contains behaviour trees, main launch files, parameters files, rviz configuration and main parameter file shared between ROS and MAES, i.e. maes_config.yaml. In order to launch a multi robot system, the maes_ros2_multi_robot_launch.py file is used. |
| ROS-TCP-Endpoint | A node relaying all messages from the ROS-TCP-Connector, which receives the ROS msgs sent inside MAES. The official repository can be found here: https://github.com/Unity-Technologies/ROS-TCP-Endpoint. We use version 0.7.0. The TCP connector inside MAES is also version 0.7.0. |
| slam_toolbox | Officially available at https://github.com/SteveMacenski/slam_toolbox/. We included it here, since the official release of 2.5.1 from September 2021 had a bug, where the odom was constantly published to /tf. The version in our repository is from the galactic branch and the latest commit has message "Backport 479 galactic" and was committed on 25. march 2022. |
Maes supports extraction of data as csv files regarding both coverage, exploration and communication. Configuring statistics gathering is done in the GlobalSettings.cs for UnityMode and in the maes_config.yaml for ROSMode. Here the path for the statistics files can be changed, statistics gathering can be enabled/disabled and the interval for saving the data of a given simulation can be changed. The csv files are created whenever a simulation finishes, if it is enabled. A simulation finishes when the end criteria from the configuration is reached.
The default path for the csv files is on the desktop inside a /MaesStatistics/ folder, that is created in the process.
The csv files can be used directly or processed and turned into tables using this python script.
Maes supports headless runs by utilizing Unity's own batch mode. These runs cannot be executed through the Unity Editor, and must be run as compiled builds.
A headless run is invoked by adding the command line argument -batchmode when executing the compiled build.
- If the "Server Build" build setting is checked, the resulting build will output CLI-like information when executing.
- If the "Server Build" build setting is left unchecked, a black window will render on the screen while the headless run is executing.
Headless runs will start simulating immediately on the "Fast as possible" speed setting (
), until the scenario queue is empty.
When in batch mode, the application will quit automatically when the scenario queue is empty.
This repository contains a bash-script for continuously logging memory usage, CPU utilization, and network activity. The script currently assumes that the network activity to be logged is happening on the docker0 interface. If you are not running anything in a container, please change the captured interface accordingly in the script. Memory use and CPU utilization is measured as system-wide measurements.
The script is especially useful when measuring whether changes made to an exploration algorithm (or the MAES-tool itself) have reduced or increased resource usage.
Open a terminal and run the script to start logging. The script checks for missing packages, and will abort if any are not found. Data is logged to separate .csv-files (values separated by whitespace), each with a name ending in network/cpu/memory. First data-entry on every line is always Unix-Epoch, so it is easier to align the data.
While the script is logging, it will prompt for entering in names of events. These can be used as "bookmarks" for interpreting the data at a later point, making it easier to determine at which epoch some event happened. The "bookmarks" are saved in a separate file.
NOTE: If you add new external dependencies to the MAES code, you must ensure that the assembly definitions for these dependencies are referenced in the CustomScriptsAssembly found in the Assets/Scripts folder. This can be done via the unity editor.
If you do not add the dependencies to this assembly file the program and the unit tests will not compile.
In order to assure functionality before any contributions some tests (unit and system tests) have been designed. For now, only the unit tests are automated.
The system test includes using both ROS and MAES, thus the entire system. The test can be used by using the following guide:
- Replace the content of maes_config.yaml with the content of maes_config_ros_system_test.yaml
- Ensure that the maes_robot_controller.py uses the default example frontier algorithm found in the main branch
- Colcon build the workspace and run it using the following command
ros2 launch maes_ros2_interface maes_ros2_multi_robot_launch.py- Run MAES
- Press play in MAES
A single robot will start exploring. The configuration has been shown to achieve about 99.9% after about 1480 ticks (2:28 seconds). The result can deviate with up to 10 seconds. If the test still yields these results after your code contribution, the system appears to be functioning correctly.
The unit tests uses the Unity framework and splits tests into two categories: Edit Mode tests and Play Mode tests.
Edit mode tests resemble traditional unit tests and run independently of the simulator. These tests can be found in the EditTests folder.
Play mode tests on the other hand run along side a unity instance of the simulator. This allows for testing of features that require the simulator to run while testing. These tests can be found in the PlayModeTests folder.
To run these test go into the unity editor -> Window -> General -> Test Runner.
Map Configuration:
| Name | Type | Meaning |
|---|---|---|
| Width | Int | Width in tiles |
| Height | Int | Height in tiles |
| Random Seed | Int | Affects the map generation |
| Border size | Int | Makes tiles up to n tiles from the border solid, i.e. not traversable |
| Scaling | Float | Scales the map, which affects the robots size relative to the map. NOTE: Sometimes has bugs |
| Random Fill Percent (cave map) | Float in ]0.0-100.0] | Determines the amount of the map that is filled with solid tiles |
| Smoothing Runs (cave map) | Int | Smoothens the map like conways game of life. i.e tiles with many solid neighbors turn solid |
| Connection Passages Width (cave map) | Int | Some rooms may not be interconnected after the smoothing runs. In this step corridors are made between the rooms. This parameter determines the width of the corridors. |
| Wall Threshold Size (cave map) | Int | All groups of wall tiles smaller than this will be made open (traversable) |
| Room Threshold Size (cave map) | Int | All groups of open tiles smaller than this will be made solid (non-traversable) |
| Max Hall Percent (building map) | Float in ]0.0-100.0] | Hall are generated until no longer possible (Minimum room side length disallows further splitting) or until this percentage is reached |
| Hall Width (building map) | Int | The width of the generated halls in tiles |
| Minimum Room Side Length (building map) | Int | No room can have a side shorter than this value. A high value results in bigger rooms |
| Door Width (building map) | Int | The width of the doors connecting two rooms |
| Door Padding (building map) | Int | Minimum distance from a door to a wall intersection. A higher value puts the more in the middle of wall sections. |
Agent Constraints:
| Name | Type | Meaning |
|---|---|---|
| Automatically Update SLAM | Bool | Disables SLAM, which disables position approximation |
| SLAM Update Interval in Ticks | Int | SLAM map and position is updated at this rate (10 ticks = 1 second) |
| SLAM Synchronize Interval in Ticks | Int | If agents are within broadcast range (also includes blocked by walls) they will synchronize maps at this rate (10 ticks = 1 second) |
| SLAM Positioning Inaccuracy | Float | An agent's actual position may differ by up to this value in both the x and y axis |
| Distribute SLAM | Bool | If true, agents will attempt to distribute their slam at every slam synchronize interval |
| Slam Ray Trace Count | Int | Optional value for setting amount of ray traces for each agent. If not set, it will automatically be calculated depending on the range of the SlamRayTraces. |
| Slam Ray Trace Range | Float | Used for ray tracing the vision of an agent. Everything within line of sight and this distance will be marked as explored |
| Environment Tag Read Range | Float | Determines as which range measured in tiles an agent can read a tag |
| Relative Move Speed | Float | Speed relative to tiles. A bigger map with bigger doors would f.x. make the agent move relatively slower, if this parameter remained the same. 1.0f is default. 10f would make the force placed on the robot 10x larger |
| Robot Relative Size | Float | A value in ]0, 1.0] that determines the size of the agent relative to a single tile in the environment |
| Sense Nearby Agents Range | Float | The range at which agents knows of other agents' presence, i.e. distance and angle to the other agent measured in tiles |
| Sense Nearby Agents Blocked by Walls | Bool | If true, agents only know of other agents' presence, if they are within line of sight |
| MinimumSignalTransmissionProbability | Float | The smallest probability value that still allows communication between robots |
| CalculateSignalTransmissionProbability | Delegate | A function that provides a value representing the probability of transmitting a message. If this value is above the MinimumSignalTransmissionProbability then the message will be transmitted, otherwise it will be discarded. This function is given the total distance travelled by the signal and the distance travelled through walls. |
| MaterialCommunication | Bool | If true, agents use attenuationDictionary for communication instead. |
| Frequency | UInt | The frequency of the communication when using MaterialCommunication |
| TransmitPower | Float | The power in Db of the sender |
| ReceiverSensitivity | Float | The minimum Db for the receiver to accept a message |
| AttenuationDictionary | Dictionary<uint, Dictionary<TileType, float>>? | the attenuations of different tile types e.g. Concrete. Only used for material communication |
In the YAML file you cannot supply a function for the message transmission probability. Instead you can supply a maximum broadcast range and a parameters that indicates whether message can pass through walls:
** YAML ONLY **
| Name | Type | Meaning |
|---|---|---|
| Broadcast Range | Float | The range at which agents can communicate measured in tiles |
| Broadcast Blocked by Walls | Bool | If true, agent communication requires line of sight |
Agent Spawn Configuration:
| Name | Type | Meaning |
|---|---|---|
| Spawn Configuration | Delegate with type: List<MonaRobot> RobotFactory(SimulationMap map, RobotSpawner spawner) | A function for spawning the agents in a specific way. Presets are available, such as "togetherAroundPoint" and "spawnInBiggestRoom". Additionally, "inHallways" is a building map type specific spawning configuration |
| Number of Agents | Int | The number of agents spawned into the map |
| Random Seed | Int | Used to provide agents with individual random seeds |
| Exploration Algorithm | Delegate with type: IExplorationAlgorithm CreateAlgorithmDelegate(int randomSeed) | A function that returns an instance of the exploration algorithm with its dependencies injected (e.g. random seed or other algorithm specific parameters) |
Global settings
MAES contains several settings that influences the behaviour of the simulation. These can be found in the GlobalSettings.cs file. An explanation can be found below
| Name | Type | Description |
|---|---|---|
| Logic Tick Delta Millis | Int | The amount of simulated milliseconds between each logic tick, i.e. running of code from an exploration algorithm |
| Physics Ticks Per Logic Update | Int | The amount of times Unity will continue to simulate physics between each logic update. Setting this to 1 will ensure that no movement occurs between two logic ticks |
| Draw Communication | Bool | If enabled, use Unity Gizmos to draw lines between communicating agents when the communicate (Can only be seen inside Unity Editor with Gizmos enabled) |
| Should Write CSV Results | Bool | If enabled, write statistics to a CSV files whenever a simulation is finished (Not when the simulator is closed manually) |
| Statistics Output Path | string | The directory where the CSV results are saved upon completing a simulation. Remember to end string with Path.DirectorySeparatorChar |
| Ticks Per Stats Snapshot | Int | The frequency of data collection from the simulation. If this value is 1 the state of the simulation is measured and saved at every logic tick |
| Populate Adjacency And Com Groups Every Tick | Bool | The adjacency matrix used inside the CommunicationManager.cs is populated lazily. Enabling this setting will make it eager. This can be useful for gathering statistics regarding communication ranges to test if agents are at any time outside communication range, as opposed to testing only when communication actually occurs. Enabling this does, however, decrease performance - sometimes significantly so |
| Ticks Before Exploration Heat Map Cold | Int | The amount of ticks that need to pass without exploration before the exploration heat map will show that cell as completely cold. |
| Ticks Before Coverage Heat Map Cold | Int | The amount of ticks that need to pass without coverage before the coverage heat map will show that cell as completely cold. |
To run the Minotaur experiments on your own, then you can use the Unity scenes in ./Assets/Scenes/ExperimentSimulations. The building- and cave maps are seperated into their own folders. The scenes can be built on their own and should output 1000 csv files with 1-9 robots, spawnapart and together. The scenes are dependent on the name as it gets split and used for the RunSimulation of ExperimentBase.
Further Scenes can be added for eventual new algorithms or configuration options by creating a new scene with the configuration as part of the name, and giving it a gameObject with SceneToExperiment as a script.
This has been done so the individual scenes can be built seperately and run with the -batchmode arg for parallelisation. The ExperimentBase can also be used on its own to run specific simulations with the optional command.
Philip Irming Holler - [email protected]
Magnus Kirkegaard Jensen - [email protected]
Malte Zoëga Andreasen - [email protected]
Rasmus Borrisholt Schmidt - [email protected]
Andreas Sebastian Sørensen - [email protected]
Thor Beregaard - [email protected]