Adaptive Bridge: A Proxy-Based Decoupling Layer for Mitigating DDS Backpressure in ROS 2
Kaushalraj Puwar, B. Thangaraju
International Institute of Information Technology Bangalore
Published September 6, 2026 · on the feed September 11, 2026 · ▲ 25 · arXiv 2608.15380
The abstract
In systems built on Robot Operating System 2 (ROS 2) and using Data Distribution Service (DDS), a single network-impaired or throttled subscriber on a RELIABLE topic can cause backpressure that degrades throughput and latency for all other subscribers, including safety-critical ones sharing the publisher, because the publisher's DDS writer can no longer accept new samples. We present Adaptive Bridge, a proxy-based layer that decouples critical subscribers from degraded or noncritical ones, thereby isolating the critical path through topic splitting and dynamic rate control. The proxy acts as a middleman and subscribes to the original topic and republishes the messages to two independent DDS writers: one RELIABLE writer for critical nodes and one BEST EFFORT writer for noncritical or degraded nodes, thus isolating the degraded nodes and safeguarding the publisher and critical nodes from backpressure. A probe-based classifier actively monitors subscriber health through sampling with hysteresis and adjusts subscriber rate limits in real time. We evaluate the system under a Gilbert-Elliott bursty wireless loss model using a reproducible Docker-based harness. The results show that using the Adaptive Bridge in our evaluation harness reduces the critical subscriber tail p95 latency from up to 15 s to 1.55 ms across all impairment severities while preserving the publisher's configured throughput.
The Brief
2:39A short podcast summary of the paper. Written and voiced by AI.
Transcript
Adaptive Bridge. A Proxy-Based Decoupling Layer for Mitigating DDS Backpressure in ROS 2. By Kaushalraj Puwar and one other, from the International Institute of Information Technology Bangalore. Published on the sixth of September. Its code repository is worked out: about fifty-five source files, with a readme and a licence.
Robot software built on ROS 2 moves its messages over the Data Distribution Service, DDS. On a topic marked RELIABLE, every subscriber has to keep up, and that creates a vulnerability. A single subscriber that is network-impaired or throttled can cause backpressure: the publisher's DDS writer can no longer accept new samples. When that happens the damage does not stay with the slow subscriber. Throughput and latency degrade for every other subscriber sharing the publisher, including the safety-critical ones. One bad wireless link on a noncritical node can end up holding up the node that must never be held up.
Adaptive Bridge answers this by putting something in between. It is a proxy-based layer that decouples critical subscribers from degraded or noncritical ones, so that the critical path is isolated. Two tools do the isolating: topic splitting, and dynamic rate control. The publisher no longer talks to everyone directly, so a struggling subscriber no longer has a line back to the publisher through which to push.
The proxy acts as a middleman. It subscribes to the original topic, then republishes the messages to two independent DDS writers. One is a RELIABLE writer, for the critical nodes. The other is a BEST EFFORT writer, for noncritical or degraded nodes. Because the degraded nodes now hang off the best-effort writer, their trouble is contained there, and both the publisher and the critical nodes are safeguarded from backpressure.
Deciding which nodes count as degraded is the job of a probe-based classifier. It actively monitors subscriber health by sampling, with hysteresis, so that a node is not flipped back and forth on every small fluctuation. It also adjusts subscriber rate limits in real time, which is the dynamic rate control half of the design.
The evaluation uses a Gilbert-Elliott model of bursty wireless loss, run inside a reproducible Docker-based harness. Across all impairment severities, the critical subscriber's p95 tail latency falls from as much as fifteen seconds to one point five five milliseconds when the Adaptive Bridge is in place. At the same time, the publisher keeps the throughput it was configured for. The authors are careful to frame this as a result within their evaluation harness.
The design choice at the centre is a separation of fates. Reliability is kept for the nodes that need it, best effort is offered to the nodes that cannot keep up, and a monitoring classifier decides in real time who belongs where. The critical path stops depending on the slowest listener on the topic.
Paper Quest · 7 chapters · about 41 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Every quest starts with a wall. Here it is a Wi-Fi hiccup that freezes a robot's safety system.
Your mobile robot publishes laser scans at 30 Hz. Local collision avoidance reads them, and so does a remote visualisation screen over Wi-Fi. Both use reliable delivery from the same writer.
The road ahead
- 1 · The problem ~7 min When readers share one reliable writer, a single slow reader on bad Wi-Fi can stall the publisher and starve a safety-critical reader.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min Stop sharing the writer: a proxy republishes each message through two independent writers, reliable for critical readers and best effort for the rest.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min The proxy reads the topic, classifies each subscriber by probing it, sends to critical ones reliably and degraded ones best effort, and throttles the latter.
- 6 · The experiment ~6 min You’ll guess: Under bursty wireless loss on one remote subscriber, what was the critical subscriber's p95 latency with the bridge in place?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.