noelex / rclnet

Modern ROS 2 client library for .NET.
MIT License
19 stars 2 forks source link
csharp dotnet rcl ros ros2

rclnet

rclnet is a fast and easy-to-use .NET wrapper over ROS 2 client library, allowing .NET applications to interact with other ROS applications.

Features

Supported ROS Features

Feature Status Additional Information
Topics N/A
Services N/A
Actions Managed implementation.
Clocks Supports external time source by setting use_sim_time to true.
CancellationTokenSources can also be configured to cancel with timeout measured by external clock.
Timers N/A
Guard Conditions N/A
Events Event handlers can be registered via SubscriptionOptions or PublisherOptions when creating the subscirption or publisher.
ROS Graph Managed implementation.
Logging Supports logging to stdout, /rosout and log files. Configurable with --ros-args.
Content Filtered Topics Available since humble.
Network Flow Endpoints Available since galactic.
Network flow endpoints of publishers and subscriptions can be retrieved via IRclPublisher.Endpoints and IRclSubscription.Endpoints property.
Unique network flow endpoints requirement can be configured when creating SubscriptionOptions and PublisherOptions.
Service Introspection Available since iron.
Parameter Service ⚠️ Supports loading parameters from command-line arguments and parameter files.
Locally declared parameters are exposed via Parameter API.
Parameter client is not implemented.
Lifecycle Nodes N/A

✅Supported ⚠️Partial support ❌Not supported ⏳In development

Supported Platforms

Supported .NET Versions:

Supported ROS 2 Distributions:

Supported Operating Systems:

Should also work on macOS but untested.

Getting Started

rclnet provides project templates to help you getting started quickly. You can install the templates using the following command:

dotnet new install Rcl.NET.Templates

Then, to create a node application:

mkdir MyNode
cd MyNode
dotnet new ros2-node

To create a message only library:

mkdir MyMessageLibrary
cd MyMessageLibrary
dotnet new ros2-msg

If you prefer creating the projects by yourself, or want to use rclnet in existing projects, refer the Installing and Generating Messages section.

Installing

Stable releases of rclnet are hosted on NuGet. You can install them using the following command:

dotnet add package Rcl.NET

For message only projects, you can install only Rosidl.Runtime without Rcl.NET:

dotnet add package Rosidl.Runtime

To generate message classes, you'll also need to install ros2cs utility:

dotnet tool install -g ros2cs

Generating Messages

rclnet does not ship with message definitions. In order to communicate with other ROS 2 nodes, you need to generate messages first.

Message definitions are .NET classes / structs, you can either include messages in a console app which runs as an ROS 2 node, or compile separately in another library.

Projects containing messages will have to meet the following requirements:

To generate messages, you also need to add a ros2cs.spec file to somewhere in the project (usually the project root). A ros2cs.spec file contains configurations such as output directory and where to find packages, see here for detailed explanations.

Assuming you've already installed the ros2cs utility, simply run the following command to generate messages:

ros2cs /path/to/ros2cs.spec

Configuring Message Generation Using Command-line Arguments

Starting from v1.3.0, ros2cs supports specifying message generation options using command-line arguments. You can configure message generation using command-line arguments in conjunction with ros2cs.spec files.

Usage:  ros2cs <OPTIONS> <SPEC_FILE>

Generates C# source code from ROS2 interface definitions for use with Rcl.NET.

OPTIONS:
    --service-introspection,
    --no-service-introspection
        Generate code for service introspection.

        This feature requires 'service_msgs' package.
        This option is disabled by default.

    --action-details,
    --no-action-details
        Generate implementation details for actions, including FeedbackMessage, SendGoal
        request/response and GetResult request/response.
        Detail messages are not neccesary for using action servers and clients as rclnet
        utilizes introspection to construct and extract Feedback / Goal and Result messages.
        Enabling 'action-details' will allow you to perform service introspection on Result
        and Goal services, or manually subscribe Feedback messages, at the cost of increased
        code size.

        This feature requires 'unique_identifier_msgs' and 'builtin_interfaces' package.
        This option is disabled by default.

    --internal,
    --no-internal
        Generate classes with internal visibility rather than public.

        This option is disabled by default.

    --from-ament-index,
    --no-ament-index
        Read packages from ament prefix directories as specified by AMENT_PREFIX_PATH.
        If this flag specified, ament prefix directories always load before all other
        'from-directory' directives.

        This option is disabled by default.

    --ignore-missing,
    --no-ignore-missing
        Ignore missing included and depedency packages, proceed with packages currently
        available anyway.

        By default, ros2cs will abort with exit code 1 if any package dependency is missing,
        unless missing packages are excluded explicitly using 'exclude' option.
        Enabling this option forces ros2cs to generate code even if some packages are missing.

        This option is disabled by default.

    -o,
    --output=OUTPUT_DIR
        Specifiy the output directory.
        A path relative to the SPEC_FILE (if not specified, relative to current directory),
        or an absolute path can be used.

    -I,
    --from-directory=INCLUDE_DIR
        Read packages from specified root package directory containing one or more packages.
        A path relative to this file or an absolute path can be used.

        This option may be specified multiple times.

    -n,
    --namespace=NAMESPACE
        Set the default root namespace of the generated code.

    -i,
    --include="PKG1 PKG2 ..."
        Generate codes for specified packages only.
        Dependencies of the specified packages will also be included recursively.
        If no include directive is specified, ros2cs will generate codes for all discovered
        packages.

        This option may be specified multiple times.

    -e,
    --exclude="PKG1 PKG2 ..."
        Explicitly exclude packages from code generation.

        This option may be specified multiple times.

    --map-namespace=PKG:NAMESPACE
        Set the root namespace of a specific package,
        e.g. '--map-name my_messages:My.Namespace' will map definitions in package 'my_messages'
        into C# namespace My.Namespace.MyMessages.
        If not specified, default root namespace is used.

        This option may be specified multiple times.

    --map-name=PKG:NAME
        Set the name of a specific package,
        e.g. 'map-name my_messages:Messages' will map definitions in package 'my_messages'
        into C# namespace Rosidl.Messages.Messages.
        If not specified, ros2cs will determine the package name automatically.

        This option may be specified multiple times.

    -v,
    --version
        Print the version of this program.

    -h,
    --help
        Print this message.

SPEC_FILE:
    Path to a ros2cs spec file containing configurations for generating source code.

    If neither OPTIONS nor SPEC_FILE is specified, ros2cs will try to use 'ros2cs.spec'
    in current directory as SPEC_FILE.

    If both OPTIONS and SPEC_FILE are specified, ros2cs will load configurations from
    SPEC_FILE first, and override configurations with options specified by OPTIONS.

Asynchronous Execution Model

Unlike rclcpp and rclpy, rclnet doesn't have the concept of executors. Each RclContext runs its own event loop for waiting on signals and dispatching callbacks, which is essentialy a single-threaded executor.

Although rclnet does not provide multi-threaded executors, it doesn't mean that you can't process messages or handle service requests using multiple threads. All communication primitives in rclnet provide both synchronous and asynchronous APIs for different needs and scenarios.

Synchronous APIs are simpler and faster if the work need to be done is simple enough, e.g. neither CPU-intensive nor needs to issue blocking calls. Asynchronous APIs, in contrast, are for scenarios where you need to perform asynchronous calls or offload blocking operations into background threads.

Take subscriptions for example, you can receive messages synchronously using IRclSubscription<T>.Subscribe, or asynchronously using IRclSubscription<T>.ReadAllAsync. Synchronous subscriptions always handle messages on the event loop. While for asynchronous subscriptions, you can choose where you'd like to process the received messages:

await foreach (var msg in sub.ReadAllAsync())
{
    // Perform asynchronous operation.
    await SomeAsyncOperation(msg);

    // Perform synchronous operation and wait for its completion without blocking the event loop.
    await Task.Run(() => SomeOffloadedSyncOperation(msg));
}

In the above example, the event loop of the RclContext is used for listening to events only. Where are the messages handled depends on the SynchronizationContext currently captured.

If there's no SynchronizationContext in use, event handling happens in background threads by default. Otherwise, the events will be handled in the captured SynchronizationContext. If you are using rclnet inside a GUI application, this usually means that the events are handled on the UI thread.

RclContexts can also have their own SynchronizationContexts, which always schedule asynchronous operations on the event loop. This is extremely helpful if you want to introduce single-threaded concurrency into your application:

await using var context = new RclContext(useSynchronizationContext: true);

...

// Enforce execution on the event loop so that we can capture its SynchronizationContext.
await context.Yield();

// All following awaits will resume on the event loop by default.
await foreach (var msg in sub.ReadAllAsync())
{
    // On event loop.
    await SomeAsyncOperation(msg);
    // On event loop.
    await Task.Run(() => {
        // On thread pool.
        SomeOffloadedSyncOperation(msg);
    });
    // On event loop.
    await Task.Yield();
    // On event loop.

    // We can also spin up multiple coroutines to run concurrently on the event loop.
    Task task1 = Coroutine1Async(msg),
         task2 = Coroutine2Async(msg);

    // Or asynchronously wait for all coroutines to complete.
    await Task.WhenAll(task1, task2);

    ...

    // The execution of current async method will stay on the event
    // loop unless we break out of the SynchronizationContext using
    // ConfigureAwait(false), or RclContext.YieldBackground().

    await AnotherAsyncOperation(msg).ConfigureAwait(false);
    // On thread pool thread.

    // We can still transition back to the event loop with context.Yield().

    await context.Yield();
    // On event loop.

    await RclContext.YieldBackground();
    // On thread pool thread.
}

As shown in the above example, besides of SynchronizationContext, you can also use RclContext.Yield, RclContext.YieldBackground and ConfigureAwait(false) to perform fine-grained control over the asynchronous exection flow.

Additional Notes about IRclWaitObject.WaitOneAsync

Timers and guard conditions created by RclContext implements IRclWaitObject interface, which allow the caller to asynchronously wait for the signal.

IRclWaitObject interface exposes the following two overloads of WaitOneAsync:

ValueTask WaitOneAsync(bool runContinuationAsynchronously, CancellationToken cancellationToken = default);
ValueTask WaitOneAsync(CancellationToken cancellationToken = default);

The latter overload simply calls another one with runContinuationAsynchronously set to true.

WaitOneAsync allows the caller to explicitly control the execution of the continuation via runContinuationAsynchronously parameter. Assuming there's no captured SynchronizationContext or TaskScheduler, when runContinuationAsynchronously is set to true, the continuation will be scheduled to execute in thread pool. And if runContinuationAsynchronously is set to false, the continuation is guaranteed to execute on the event loop.

However, when a SynchronizationContext or TaskScheduler is captured, the continuation of the call to WaitOneAsync will always execute in the captured context, regardless of the value of runContinuationAsynchronously.

Since context capture can be suppressed by calling ConfigureAwait with continueOnCapturedContext set to false, execution of the continuation can be precisely controlled using runContinuationAsynchronously in conjunction with continueOnCapturedContext.

runContinuationAsynchronously continueOnCapturedContext Continuation Execution
true true Captured SynchronizationContext or TaskScheduler if any, thread pool otherwise
true false Thread pool
false true Captured SynchronizationContext or TaskScheduler if any, event loop otherwise
false false Event loop

Building and Running Examples

Install dependencies

The following instruction assumes that you've already installed ROS 2 foxy or humble in your system.

You'll need .NET 8.0 SDK to build and run the examples, see instructions here.

Make sure you have all dependencies installed by running:

rosdep install -i --from-paths examples

Run with dotnet run

Now you can run example projects using dotnet run, e.g.

dotnet run examples/turtle_rotate

Run with ros2 run

Or you can build and install examples as colcon packages:

colcon build --executor sequential --merge-install --paths examples/*
source install/setup.bash

To run an example node, use ros2 run, e.g.

ros2 run graph_monitor graph_monitor

Showcase

Subscribing

await using var ctx = new RclContext(args);
using var node = ctx.CreateNode("hellow_world");
using var sub = node.CreateSubscription<Twist>("/cmd_vel");
await foreach (Twist msg in sub.ReadAllAsync())
{
    ...
}

Publishing

using var pub = node.CreatePublisher<Vector3>("/vec");
pub.Publish(new Vector3(x: 1, y: 2, z: 3));

Handling Service Calls

using var server = node.CreateService<
    EmptyService,
    EmptyServiceRequest,
    EmptyServiceResponse>("/vec",
        (request, state) =>
        {
            return new EmptyServiceResponse();
        });
await Task.Delay(-1);

Calling Services

using var client = node.CreateClient<
    EmptyService,
    EmptyServiceRequest,
    EmptyServiceResponse>("/vec");
await client.InvokeAsync(new EmptyServiceRequest());

Monitoring ROS Graph Changes

node.Graph
    .OfType<NodeAppearedEvent>()
    .Subscribe(x =>
    {
        Console.WriteLine($"Node {x.Node.Name} is online.");
    });

await node.Graph.WaitForServiceServerAsync("/my/service");

Calling Action Servers

using var client = node.CreateActionClient<
    SpinAction,
    SpinActionGoal,
    SpinActionResult,
    SpinActionFeedback>("/spin");

using var goal = await client.SendGoalAsync(
        new SpinActionGoal(targetYaw: Math.PI));

await foreach (var feedback in goal.ReadFeedbacksAsync())
{
    Console.WriteLine(feedback.AngularDistanceTraveled);
}

var result = await goal.GetResultAsync();

Zero (Managed Heap) Allocation APIs

using var sub = node.CreateNativeSubscription<Twist>("/cmd_vel");
await foreach (RosMessageBuffer msg in sub.ReadAllAsync())
{
    using (msg) ProcessMessage(msg);

    static void ProcessMessage(RosMessageBuffer buffer)
    {
        ref var twist = ref buffer.AsRef<Twist.Priv>();
        ...
    }
}