SWARM Modules & Algorithms | SWARM Simulation Firmware Documentation

This page describes all available algorithms in the SWARM Developer platform, including available parameters.

SWARM RDS Modules

The SWARM RDS Platform is a highly modular system, allowing for the easy integration of new algorithms and solutions. The following is a list of the current modules available in the SWARM RDS Platform. Each module has it's own unique set of parameters and requirements, so please ensure you read the documentation for each module before attempting to use it.

Each Module runs within it's own Python process, with a unique messaging system designed to ensure that data is passed between modules in a timely and efficient manner. This allows for the easy integration of new algorithms and solutions, as well as the ability to run multiple algorithms at the same time.

We do all of the heavy lifting, including running the Module, publishing and subscribing to data, and ensuring that the logging, data clean up and more is handled properly. For more information on how to manage data, please view: Publish and Subscribe

ObstacleAvoidance

SensorDataCollector

The SensorDataCollector module is an automated node that will build and collect data for all supported sensors that are attached to the agent. The data is collected at the prescribed "PublishingRate". All data is output in the selected output type in the module settings.

Parameters

Parameter	Data Type	Description	Available Options
OutputType	String	The type of data file that should be created with the collected data!	csv

Example Settings Input

"SensorDataCollector": {
    "Parameters": {
        "OutputType": "csv"
    },
    "Publishes": [
    ],
    "Subscribes": [
    ]
},

Parameter	Data Type	Description	Available Options
algorithm_type	String	The type of CV algorithm	Detection Classification Segmentation
model_name	String	The name of the Wrapper Python class. This must be the Exact name of the class. Furthermore, the file that this class is contained in must also be this name exactly.	Any
weight_file_name	String	The name of the weights file to be loaded into the state_dict of the model. Must be in the weights folder	Any
output_type	String	How the output of the model should be displayed	stream (Local Only - view the output live) images - save each frame with bounding boxes video - save a video with bounding boxes
camera_name	String	The name of the camera that is being used to generate input images.	Any - must be in the "Cameras" section of the agent with this module
model_folder_name	String	The name of the folder where the model is stored. Must be the same as the "Model" parameter of the module.	Any - must be in the "Cameras" section of the agent with this module
model_parameters	dict	The input parameters to your network. These are the arguements provided to the model wrapper class, allowing you to change different parameters of your network as needed.	Any - these parameters must be named exactly as listed in the wrapper class

Parameter

Data Type

Description

Available Options

algorithm_type

String

The type of CV algorithm

Detection
Classification
Segmentation

model_name

String

The name of the Wrapper Python class. This must be the Exact name of the class. Furthermore, the file that this class is contained in must also be this name exactly.

Any

weight_file_name

String

The name of the weights file to be loaded into the state_dict of the model. Must be in the weights folder

Any

output_type

String

How the output of the model should be displayed

stream (Local Only - view the output live)
images - save each frame with bounding boxes
video - save a video with bounding boxes

camera_name

String

The name of the camera that is being used to generate input images.

Any - must be in the "Cameras" section of the agent with this module

model_folder_name

String

The name of the folder where the model is stored. Must be the same as the "Model" parameter of the module.

Any - must be in the "Cameras" section of the agent with this module

model_parameters

dict

The input parameters to your network. These are the arguements provided to the model wrapper class, allowing you to change different parameters of your network as needed.

Any - these parameters must be named exactly as listed in the wrapper class

"TorchPerception": { "Algorithm": { "Level": 1, "States": [], "Parameters": { "algorithm_type": "Detection", "model_name": "SSDWrapper", "weight_file_name": "ssd_300_VOC0712.pth", "output_type": "stream", "camera_name": "Camera1", "model_folder_name": "SSD", "model_parameters": { "network_parameters": {}, "input_image_size": [ 480, 640 ] } }, "InputArgs": [ "Image" ], "ClassName": "TorchCustomModel", "ReturnValues": [ "Detection" ] }, "Parameters": { "Model": "SSD" }, "Publishes": [ "Detection" ], "Subscribes": [ "AgentState", { "Image": "Camera1" } ] },

Parameter	Data Type	Available Options	Available Options
algorithm_type	String	The type of CV algorithm	Detection Classification Segmentation
model_name	String	The name of the model to use from the Torch Hub repository	yolov5s - must be used with repo 'ultralytics/yolov5' More supported models coming soon!
output_type	String	How the output of the model should be displayed	stream (Local Only - view the output live) images - save each frame with bounding boxes video - save a video with bounding boxes
camera_name	String	The name of the camera that is being used to generate input images.	Any - must be in the "Cameras" section of the agent with this module

Parameter

Data Type

Available Options

algorithm_type

String

The type of CV algorithm

Detection
Classification
Segmentation

model_name

String

The name of the model to use from the Torch Hub repository

yolov5s - must be used with repo 'ultralytics/yolov5'
More supported models coming soon!

output_type

String

How the output of the model should be displayed

stream (Local Only - view the output live)
images - save each frame with bounding boxes
video - save a video with bounding boxes

camera_name

String

The name of the camera that is being used to generate input images.

Any - must be in the "Cameras" section of the agent with this module

"TorchPerception": { "Algorithm": { "Level": 1, "States": [], "Parameters": { "model_type": "Detection", "model_name": "yolov5s", "repo_name": "ultralytics/yolov5", "output_type": "stream", "camera_name": "Camera1" }, "InputArgs": [ "Image" ], "ClassName": "TorchHubModel", "ReturnValues": [ "Detection" ] }, "Parameters": {}, "Publishes": [ "Detection" ], "Subscribes": [ "AgentState", { "Image": "Camera1" } ] },