PyZeROS

Python-only ROS 2. No rclpy, no ROS installation, no message compilation. Just pip install and go.

Built on Zenoh, asyncio-for-robotics, and ros2-pyterfaces.

import asyncio, pyzeros
import asyncio_for_robotics as afor
from ros2_pyterfaces.cyclone.all_msgs import String

@afor.scoped
async def main():
    sub = pyzeros.Sub(String, "chatter")
    async for msg in sub.listen_reliable():
        print(msg.data)

with pyzeros.auto_context(node="listener", namespace="/demo"):
    asyncio.run(main())

Features:

• Topics and services, fully interoperable with ROS 2 nodes
• asyncio execution model. No callbacks, no spinners, just Python
• Define ROS messages in Python with ros2_pyterfaces
• Resource lifecycle via sessions and scopes

Note

Experimental. Actions and zero-copy are planned.

---

ROS 2 interop

PyZeROS talks to ROS 2 through Zenoh. The ROS 2 side must use rmw_zenoh_cpp:

export RMW_IMPLEMENTATION=rmw_zenoh_cpp

Both sides must share the same ROS_DOMAIN_ID (defaults to 0) and be on the same Zenoh network. In practice this means running a Zenoh router and configuring both sides as clients to it, or using a peer-to-peer Zenoh config for DDS-like multicast discovery.

See the rmw_zenoh docs for router setup and configuration.

Install

PyZeROS is a normal Python package. Add it as a dependency to your project with your tool of choice. No colcon, no workspace, no overlay.

pip install git+https://github.com/2lian/pyzeros2

Note

pip install pyzeros is not yet available. For now, we prioritise install from source using pixi.

From source

git clone https://github.com/2lian/pyzeros2
cd pyzeros2
pixi install
pixi run router   # start a local Zenoh router
pixi run example  # run the minimal subscriber

---

Tutorial

This mirrors the official ROS 2 tutorials, but without C++, build systems, or boilerplate.

1. Publisher and subscriber

The ROS 2 tutorial for this is here. Below is the PyZeROS equivalent.

Publisher:

import asyncio
import pyzeros
import asyncio_for_robotics as afor
from ros2_pyterfaces.cyclone.all_msgs import String

@afor.scoped
async def main():
    pub = pyzeros.Pub(String, "chatter")
    counter = 0
    async for _ in afor.Rate(2).listen():
        pub.publish(String(data=f"Hello World: {counter}"))
        print(f"Publishing: Hello World: {counter}")
        counter += 1

with pyzeros.auto_context(node="talker", namespace="/demo"):
    asyncio.run(main())

Subscriber:

import asyncio
import pyzeros
import asyncio_for_robotics as afor
from ros2_pyterfaces.cyclone.all_msgs import String

@afor.scoped
async def main():
    sub = pyzeros.Sub(String, "chatter")
    async for msg in sub.listen_reliable():
        print(f"I heard: {msg.data}")

with pyzeros.auto_context(node="listener", namespace="/demo"):
    asyncio.run(main())

That's it. No rclpy.init(), no spin(), no executor. The async for loop is the executor.

auto_context creates a session (Zenoh transport + ROS node identity) and makes it the default for everything inside. @afor.scoped ensures all resources created inside are cleaned up when the function returns. See asyncio-for-robotics for details on scopes.

2. Service and client

The ROS 2 tutorial for this is here. Below is the PyZeROS equivalent.

Server:

import asyncio
import pyzeros
import asyncio_for_robotics as afor
from ros2_pyterfaces.cyclone.all_srvs import AddTwoInts

@afor.scoped
async def main():
    server = pyzeros.Server(AddTwoInts, "add_two_ints")
    print("Service ready.")
    async for responder in server.listen_reliable():
        result = responder.request.a + responder.request.b
        responder.response.sum = result
        responder.send()
        print(f"{responder.request.a} + {responder.request.b} = {result}")

with pyzeros.auto_context(node="add_server", namespace="/demo"):
    asyncio.run(main())

Client:

import asyncio
import pyzeros
import asyncio_for_robotics as afor
from ros2_pyterfaces.cyclone.all_srvs import AddTwoInts

@afor.scoped
async def main():
    client = pyzeros.Client(AddTwoInts, "add_two_ints")
    await client.wait_for_service()
    response = await client.call_async(AddTwoInts.Request(a=2, b=3))
    print(f"Result: {response.sum}")

with pyzeros.auto_context(node="add_client", namespace="/demo"):
    asyncio.run(main())

Services follow the same async for pattern as topics. The server yields Responder objects: read responder.request, fill responder.response, call responder.send().

3. Custom messages

The ROS 2 tutorial for this is here. In ROS 2 this involves .msg files, CMake, and colcon build. In PyZeROS, it's a dataclass.

ros2_pyterfaces provides two backends for message definitions:

Backend	Import	Speed	Compatibility
cyclone	ros2_pyterfaces.cyclone	Good	Full ROS 2 interop
cydr	ros2_pyterfaces.cydr	Faster	Slightly less compatible with edge cases

Both backends ship pre-built standard messages (all_msgs, all_srvs) and let you define your own.

Important

For ROS 2 interop, the typename and field names must match the ROS message definition exactly.

Defining a message:

from dataclasses import dataclass, field
from ros2_pyterfaces.cyclone import idl, all_msgs

@dataclass
class MyStatus(idl.IdlStruct, typename="my_package/msg/MyStatus"):
    header: all_msgs.Header = field(default_factory=all_msgs.Header)
    temperature: idl.types.float64 = 0.0
    labels: idl.types.sequence[str] = field(default_factory=list)
    active: bool = False

Use it like any other message:

pub = pyzeros.Pub(MyStatus, "status")
pub.publish(MyStatus(temperature=36.5, labels=["sensor_a"], active=True))

Defining a service:

from dataclasses import dataclass
from ros2_pyterfaces.cyclone.idl import IdlStruct, make_idl_service

@dataclass
class Request(IdlStruct, typename="my_package/srv/Calibrate_Request"):
    target: str = ""

@dataclass
class Response(IdlStruct, typename="my_package/srv/Calibrate_Response"):
    success: bool = False

Calibrate = make_idl_service(Request, Response)

See ros2_pyterfaces for the full type system.

---

Putting it together

A more realistic node combining topics and services with proper resource management:

import asyncio
from contextlib import suppress

import asyncio_for_robotics as afor
import pyzeros
from ros2_pyterfaces.cyclone.all_msgs import String
from ros2_pyterfaces.cyclone.all_srvs import Trigger

@afor.scoped
async def main():
    tg = afor.Scope.current().task_group
    tg.create_task(publisher())
    tg.create_task(listener())
    tg.create_task(serve_trigger())
    await asyncio.Future()

@afor.scoped
async def publisher():
    pub = pyzeros.Pub(String, "heartbeat")
    counter = 0
    async for _ in afor.Rate(1).listen():
        pub.publish(String(data=f"alive #{counter}"))
        counter += 1

async def listener():
    sub = pyzeros.Sub(String, "commands")
    async for msg in sub.listen_reliable():
        print(f"Command: {msg.data}")

async def serve_trigger():
    server = pyzeros.Server(Trigger, "reset")
    async for responder in server.listen_reliable():
        print("Reset triggered!")
        responder.response.success = True
        responder.response.message = "done"
        responder.send()

if __name__ == "__main__":
    with pyzeros.auto_context(node="my_robot", namespace="/robot"):
        with suppress(KeyboardInterrupt):
            asyncio.run(main())

What this gives you over standard ROS 2 Python:

• Scoped cleanup: @afor.scoped closes all publishers, subscribers, servers, and rates when the function exits. No dangling resources, no manual destroy_* calls.
• TaskGroup structure: Tasks run concurrently inside the scope's TaskGroup. If one crashes, the others are cancelled and the error propagates cleanly. In standard ROS 2, a crashed callback silently dies.
• Session resolution: auto_context binds a session (node identity + transport) for the block. Every Pub, Sub, Client, Server created inside auto-resolves to that session. No passing self.node around.
• Thread safety by default: Everything runs on one asyncio event loop. No GIL juggling, no executor threads, no callback reentrancy bugs.

For the scope and session system in detail, see asyncio-for-robotics.

---

Examples

Note

Those examples are meant to run out-of-the-box with an installation from source using pixi. With just a pip install you'll need to setup your environment and ROS yourself.

Examples live under pyzeros.examples.*:

Example	Run	Description
example.py	pixi run example	Minimal subscriber
basic_usage.py	pixi run python -m pyzeros.examples.basic_usage	Repeater with custom message
demo.py	pixi run demo	Ring of async tasks
custom_msgs.py	pixi run python -m pyzeros.examples.custom_msgs	Python-defined JointState

Inspect from the ROS 2 side:

pixi run -e ros ros2 topic list
pixi run -e ros ros2 topic echo /demo/chatter std_msgs/msg/String