The mqtt_client package provides a ROS nodelet that enables connected ROS-based devices or robots to exchange ROS messages via an MQTT broker using the MQTT protocol. This works generically for arbitrary ROS message types. The mqtt_client can also exchange primitive messages with MQTT clients running on devices not based on ROS.
- Installation
- Usage
- Primitive Messages
- Latency Computation
- Package Summary
- How It Works
- Acknowledgements
The mqtt_client package is released as an official ROS Noetic package and can easily be installed via a package manager.
sudo apt install ros-noetic-mqtt-client
If you would like to install mqtt_client from source, simply clone this repository into your ROS workspace. All dependencies that are listed in the ROS package.xml
can be installed using rosdep.
# mqtt_client$
rosdep install -r --ignore-src --from-paths .
The mqtt_client can be easily integrated into an existing ROS-based system. Below, you first find a quick start guide to test the mqtt_client on a single machine. Then, more details are presented on how to launch and configure it in more complex applications.
Follow these steps to quickly launch a working mqtt_client that is sending ROS messages via an MQTT broker to itself.
It is assumed that an MQTT broker (such as Mosquitto) is running on localhost:1883
.
For this demo, you may easily launch Mosquitto with its default configuration using Docker.
docker run --rm --network host --name mosquitto eclipse-mosquitto
The mqtt_client is best configured with a ROS parameter yaml file. The configuration shown below (also see params.yaml
) allows an exchange of messages as follows:
- ROS messages received locally on ROS topic
/ping/ros
are sent to the broker on MQTT topicpingpong/ros
; - MQTT messages received from the broker on MQTT topic
pingpong/ros
are published locally on ROS topic/pong/ros
; - primitive ROS messages received locally on ROS topic
/ping/primitive
are sent as primitive (string) messages to the broker on MQTT topicpingpong/primitive
; - MQTT messages received from the broker on MQTT topic
pingpong/primitive
are published locally as primitive ROS messages on ROS topic/pong/primitive
.
broker:
host: localhost
port: 1883
bridge:
ros2mqtt:
- ros_topic: /ping/ros
mqtt_topic: pingpong/ros
- ros_topic: /ping/primitive
mqtt_topic: pingpong/primitive
primitive: true
mqtt2ros:
- mqtt_topic: pingpong/ros
ros_topic: /pong/ros
- mqtt_topic: pingpong/primitive
ros_topic: /pong/primitive
primitive: true
After building your ROS workspace, launch the mqtt_client nodelet with the pre-configured demo parameters using roslaunch, which should yield the following output.
roslaunch mqtt_client standalone.launch
[ WARN] [1665575657.358869079]: Parameter 'broker/tls/enabled' not set, defaulting to '0'
[ WARN] [1665575657.359798329]: Parameter 'client/id' not set, defaulting to ''
[ WARN] [1665575657.359810889]: Client buffer can not be enabled when client ID is empty
[ WARN] [1665575657.360300703]: Parameter 'client/clean_session' not set, defaulting to '1'
[ WARN] [1665575657.360576344]: Parameter 'client/keep_alive_interval' not set, defaulting to '60.000000'
[ WARN] [1665575657.360847295]: Parameter 'client/max_inflight' not set, defaulting to '65535'
[ INFO] [1665575657.361281461]: Bridging ROS topic '/ping/ros' to MQTT topic 'pingpong/ros'
[ INFO] [1665575657.361303380]: Bridging primitive ROS topic '/ping/primitive' to MQTT topic 'pingpong/primitive'
[ INFO] [1665575657.361352809]: Bridging MQTT topic 'pingpong/ros' to ROS topic '/pong/ros'
[ INFO] [1665575657.361370558]: Bridging MQTT topic 'pingpong/primitive' to primitive ROS topic '/pong/primitive'
[ INFO] [1665575657.362153083]: Connecting to broker at 'tcp://localhost:1883' ...
[ INFO] [1665575657.462622065]: Connected to broker at 'tcp://localhost:1883'
Note that the mqtt_client successfully connected to the broker and also echoed which ROS/MQTT topics are being bridged. For testing the communication between mqtt_client, itself, and other MQTT clients, open five new terminals.
In order to test the communication among mqtt_clients, publish any ROS message on ROS topic /ping/ros
and wait for a response on ROS topic /pong/ros
.
# 1st terminal: listen for ROS messages on /pong/ros
rostopic echo /pong/ros
# 2nd terminal: publish ROS message to /ping/ros
rostopic pub -r 1 /ping/ros std_msgs/String "Hello MQTT!"
In order to test the communication between mqtt_client and other MQTT clients, publish a primitive ROS message on ROS topic /ping/primitive
, directly publish a primitive MQTT message on MQTT topic pingpong/primitive
and wait for responses on ROS topic /pong/primitive
.
# 3rd terminal: listen for primitive ROS messages on /pong/primitive
rostopic echo /pong/primitive
# 4th terminal: publish primitive ROS message to /ping/primitive
rostopic pub -r 1 /ping/primitive std_msgs/Int32 42
# 5th terminal: publish primitive MQTT message to pingpong/primitive
docker run --rm --network host eclipse-mosquitto mosquitto_pub -h localhost -t "pingpong/primitive" --repeat 20 --repeat-delay 1 -m 69
If everything works as expected, the second terminal should print a message at 1Hz, while the third terminal should print two different messages at 1Hz.
You can start the mqtt_client nodelet in a standalone nodelet manager with:
roslaunch mqtt_client standalone.launch
This will automatically load the provided demo params.yaml to the ROS parameter server. If you wish to load your custom configuration file, simply pass params_file
.
roslaunch mqtt_client standalone.launch params_file:="</PATH/TO/PARAMS.YAML>"
You can also disable parameter loading altogether by passing load_params:=false
. In this case, make sure to populate the ROS parameter server accordingly with other means.
roslaunch mqtt_client standalone.launch load_params:=false
In order to exploit the benefits of mqtt_client being a nodelet, load the nodelet to your own nodelet manager shared with other nodelets.
All available ROS parameters supported by the mqtt_client and their default values (in []
) are listed in the following.
broker:
host: # [localhost] IP address or hostname of the machine running the MQTT broker
port: # [1883] port the MQTT broker is listening on
user: # username used for authenticating to the broker (if empty, will try to connect anonymously)
pass: # password used for authenticating to the broker
tls:
enabled: # [false] whether to connect via SSL/TLS
ca_certificate: # [/etc/ssl/certs/ca-certificates.crt] CA certificate file trusted by client (relative to ROS_HOME)
client:
id: # unique ID used to identify the client (broker may allow empty ID and automatically generate one)
buffer:
size: # [0] maximum number of messages buffered by the bridge when not connected to broker (only available if client ID is not empty)
directory: # [buffer] directory used to buffer messages when not connected to broker (relative to ROS_HOME)
last_will:
topic: # topic used for this client's last-will message (no last will, if not specified)
message: # [offline] last-will message
qos: # [0] QoS value for last-will message
retained: # [false] whether to retain last-will message
clean_session: # [true] whether to use a clean session for this client
keep_alive_interval: # [60.0] keep-alive interval in seconds
max_inflight: # [65535] maximum number of inflight messages
tls:
certificate: # client certificate file (only needed if broker requires client certificates; relative to ROS_HOME)
key: # client private key file (relative to ROS_HOME)
password: # client private key password
bridge:
ros2mqtt: # array specifying which ROS topics to map to which MQTT topics
- ros_topic: # ROS topic whose messages are transformed to MQTT messages
mqtt_topic: # MQTT topic on which the corresponding ROS messages are sent to the broker
primitive: # [false] whether to publish as primitive message
inject_timestamp: # [false] whether to attach a timestamp to a ROS2MQTT payload (for latency computation on receiver side)
advanced:
ros:
queue_size: # [1] ROS subscriber queue size
mqtt:
qos: # [0] MQTT QoS value
retained: # [false] whether to retain MQTT message
mqtt2ros: # array specifying which MQTT topics to map to which ROS topics
- mqtt_topic: # MQTT topic on which messages are received from the broker
ros_topic: # ROS topic on which corresponding MQTT messages are published
primitive: # [false] whether to publish as primitive message (if coming from non-ROS MQTT client)
advanced:
mqtt:
qos: # [0] MQTT QoS value
ros:
queue_size: # [1] ROS publisher queue size
latched: # [false] whether to latch ROS message
As seen in the Quick Start, the mqtt_client can not only exchange arbitrary ROS messages with other mqtt_clients, but it can also exchange primitive message data with other non-mqtt_client MQTT clients. This allows ROS-based devices to exchange primitive messages with devices not based on ROS. The primitive
parameter can be set for both ROS-to-MQTT (bridge/ros2mqtt
) and for MQTT-to-ROS (bridge/mqtt2ros
) transmissions.
If a ROS-to-MQTT transmission is configured as primitive
, the ROS message is simply serialized to a string representation, without providing any information on the underlying ROS message type via MQTT. If the ROS message type is one of the supported primitive ROS message types, the encapsulating ROS message components are also removed, s.t. only the raw data is published as a string. The supported primitive ROS message types are std_msgs/String
, std_msgs/Bool
, std_msgs/Char
, std_msgs/UInt8
, std_msgs/UInt16
, std_msgs/UInt32
, std_msgs/UInt64
, std_msgs/Int8
, std_msgs/Int16
, std_msgs/Int32
, std_msgs/Int64
, std_msgs/Float32
, std_msgs/Float32
.
If an MQTT-to-ROS transmission is configured as primitive
, the MQTT message is interpreted and published as a primitive data type, if possible. The message is probed in the following order: bool
(std_msgs/Bool
), int
(std_msgs/Int32
), float
(std_msgs/Float32
), string
(std_msgs/String
).
The mqtt_client provides built-in functionality to measure the latency of transferring a ROS message via an MQTT broker back to ROS. Note that this functionality is only available for non-primitive messages (see Primitive Messages). To this end, the sending client injects the current timestamp into the MQTT message. The receiving client can then compute the latency between message reception time and the injected timestamp. Naturally, this is only accurate to the level of synchronization between clocks on sending and receiving machine.
In order to inject the current timestamp into outgoing MQTT messages, the parameter inject_timestamp
has to be set for the corresponding bridge/ros2mqtt
entry. The receiving mqtt_client will then automatically publish the measured latency in seconds as a ROS std_msgs/Float64
message on topic /<mqtt_client_name>/latencies/<mqtt2ros/ros_topic>
.
These latencies can be printed easily with rostopic echo
rostopic echo --clear /<mqtt_client_name>/latencies/<mqtt2ros/ros_topic>/data
or plotted with rqt_plot:
rqt_plot /<mqtt_client_name>/latencies/<mqtt2ros/ros_topic>/data
This short package summary documents the package in line with the ROS Wiki Style Guide.
Enables connected ROS-based devices or robots to exchange ROS messages via an MQTT broker using the MQTT protocol.
<bridge/ros2mqtt[*]/ros_topic>
(topic_tools/ShapeShifter
)
ROS topic whose messages are transformed to MQTT messages and sent to the MQTT broker. May have arbitrary ROS message type.
<bridge/mqtt2ros[*]/ros_topic>
(topic_tools/ShapeShifter
)
ROS topic on which MQTT messages received from the MQTT broker are published. May have arbitrary ROS message type.~/latencies/<bridge/mqtt2ros[*]/ros_topic>
(std_msgs/Float64
)
Latencies measured on the message transfer to<bridge/mqtt2ros[*]/ros_topic>
are published here, if the received messages have a timestamp injected (see Latency Computation).
is_connected
(mqtt_client/IsConnected
)
Returns whether the client is connected to the MQTT broker.
See Configuration.
The mqtt_client is able to bridge ROS messages of arbitrary message type to an MQTT broker. To this end, it needs to employ generic ROS subscribers and publishers, which only take shape at runtime.
These generic ROS subscribers and publishers are realized through topic_tools::ShapeShifter. For each pair of ros_topic
and mqtt_topic
specified under bridge/ros2mqtt/
, a ROS subscriber is setup with the following callback signature:
void ros2mqtt(topic_tools::ShapeShifter::ConstPtr&, std::string&)
Inside the callback, the generic messages received on the ros_topic
are serialized using ros::serialization. The serialized form is then ready to be sent to the MQTT broker on the specified mqtt_topic
.
Upon retrieval of an MQTT message, it is republished as a ROS message on the ROS network. To this end, topic_tools::ShapeShifter::morph is used to have the ShapeShifter publisher take the shape of the specific ROS message type.
The required metainformation on the ROS message type can however only be extracted in the ROS subscriber callback of the publishing mqtt_client with calls to topic_tools::ShapeShifter::getMD5Sum, topic_tools::ShapeShifter::getDataType, and topic_tools::ShapeShifter::getMessageDefinition. These attributes are wrapped in a ROS message of custom type mqtt_client::RosMsgType, serialized using ros::serialization and also shared via the MQTT broker on a special topic.
When an mqtt_client receives such ROS message type metainformation, it configures the corresponding ROS ShapeShifter publisher using topic_tools::ShapeShifter::morph.
The mqtt_client also provides functionality to measure the latency of transferring a ROS message via an MQTT broker back to ROS. To this end, the sending client injects the current timestamp into the MQTT message. The receiving client can then compute the latency between message reception time and the injected timestamp. Since injection of the timestamp is optional, an extra bit of information is needed for the receiver to correctly decode the MQTT message. Therefore, the first entry in the std::vector<uint8>
message buffer is used to indicate whether the message includes an injected timestamp. The resulting std::vector<uint8>
payload takes on one of the following forms:
[ 1 | ... serialized timestamp ... | ... serialized ROS messsage ...]
[ 0 | ... serialized ROS messsage ...]
To summarize, the dataflow is as follows:
- a ROS message of arbitrary type is received on ROS topic
<ros2mqtt_ros_topic>
and passed to the generic callback- ROS message type information is extracted and wrapped as a
RosMsgType
- ROS message type information is serialized and sent via the MQTT broker on MQTT topic
mqtt_client/ros_msg_type/<ros2mqtt_mqtt_topic>
- the actual ROS message is serialized
- if
inject_timestamp
, the current timestamp is serialized and concatenated with the message - an integer is added to the message's head indicating whether a timestamp was injected
- the actual MQTT message is sent via the MQTT broker on MQTT topic
<ros2mqtt_mqtt_topic>
- ROS message type information is extracted and wrapped as a
- an MQTT message containing the ROS message type information is received on MQTT topic
mqtt_client/ros_msg_type/<ros2mqtt_mqtt_topic>
- message type information is extracted and the ShapeShifter ROS publisher is configured
- an MQTT message containing the actual ROS message is received
- depending on the first element of the message, it is decoded into the serialized ROS message and the serialized timestamp
- if the message contained a timestamp, the latency is computed and published on ROS topic
~/latencies/<mqtt2ros_ros_topic>
- the serialized ROS message is published using the ShapeShifter on ROS topic
<mqtt2ros_ros_topic>
This research is accomplished within the projects 6GEM (FKZ 16KISK036K) and UNICARagil (FKZ 16EMO0284K). We acknowledge the financial support for the projects by the Federal Ministry of Education and Research of Germany (BMBF).