In a benchmark of ROS2 communication over a wide-area network, ROS2 Connect reported the lowest latency across the tested configurations. The paper also reports higher stability and improved scalability under concurrent load than eProsima's DDS Router, rosbridge and Eclipse Zenoh. The comparison was made in one described real-WAN evaluation.
The proposal describes ROS2 Connect as a WebSocket-based framework for transparent and secure ROS2 interaction over routed WANs without changes to network infrastructure or DDS configurations. Its client-server design also sets rules for which topics, services and actions may be relayed.
A single real-WAN route
The test connected a residential client to a university-hosted server over a real WAN. A publicly reachable Apache reverse proxy was part of the setup, alongside an approximately 18 Mbit/s uplink and a 58 Mbit/s downlink. The software environment was Ubuntu 24.04 with ROS2 Jazzy Jalisco.
Researchers measured round-trip time, or RTT, the time for a serialized message to travel to the other side and return. General payload tests ran from 12 bytes to 500 kilobytes. Concurrent-load tests used message sizes from 12 bytes to 100 kilobytes and one to ten parallel topics. Each configuration used 1,000 measurements.
ROS2 Connect was tested against eProsima's DDS Router, rosbridge's WebSocket interface and Eclipse Zenoh under identical network conditions.
The connection uses a server and client
ROS2 Connect is implemented as a ROS2 package with complementary server and client nodes. The server communicates with local robotic components through native DDS, while the client uses the WebSocket connection to reach the remote side.
Authentication takes place at the application layer, and regular communication is gated until the client's credentials are verified. Authorization is configuration-driven: only explicitly specified topics, services and actions may be relayed, while unspecified access or publication is rejected.
Where the latency gap appeared
Across the tested configurations, ROS2 Connect had the lowest latency. DDS Router degraded earlier and could not reliably transmit some very large payloads. The analysis does not report the threshold at which that failure occurred.
rosbridge recorded the highest latency. Zenoh was close to ROS2 Connect, with a reported difference of approximately 1 ms for smaller payloads and up to 16 ms for larger ones; ROS2 Connect was lower in both comparisons.
The timing was less variable for ROS2 Connect than for DDS Router and rosbridge, while Zenoh showed comparable variability. The report does not supply a formal variability metric or interval.
When several topics ran at once, small-message latency stayed nearly constant up to ten topics. Large-message latency rose approximately linearly with topic count. The paper attributes that pattern to bandwidth limits and to WebSockets carrying the traffic through a single ordered TCP stream.
The result stays tied to the test
The evidence covers the described WAN path and the tested payload and topic ranges. The authors frame the findings as a comparison under that setup, not as universal superiority.
Only representative results are presented because of space constraints, with more extensive evaluation and implementation details provided in the accompanying master's thesis.
The code is available
The implementation is publicly available in a GitHub repository and released under the Mozilla Public License 2.0.
The front matter places the work in the 8th International Workshop on Robotics Software Engineering, known as RoSE'26, held in Vienna on June 1, 2026. The Virtual University of Bavaria provided continuous support since 2005, and a grant partially funded the implementation.
Paper data and sources
Original title: ROS2 Connect: A new ROS2 over WAN Solution
Authors: Daniel Schott, Lakshminarasimhan Srinivasan, Christian Herrmann, Andreas Nüchter
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: Not available
Original paper · Full text