SM2: Towards Reliable and Scalable Robotics Middleware Beyond ROS2

Modern robotic systems increasingly rely on modular software frameworks such as Robot Operating System 2 (ROS2) to support real-time, high-throughput communication across distributed components. However, as the demand for specialized hardware capabilities (e.g., GPUs or FPGAs for inference) increases, delivering self-contained robots equipped with all the necessary components is becoming increasingly challenging. The simplest and most adopted solution is to rely on dedicated edge devices with all the necessary components to perform offloading of (part of) the computation. In this regard, ROS2's reliance on Data Distribution Service (DDS) introduces significant performance degradation when networked communication is involved, often impacting even local intra-process exchanges under adverse network conditions. To address this limitation, we present Scalable Middleware 2 (SM2), a lightweight and deterministic middleware for intra- and inter-robot communication. SM2 separates local inter-process communication from network transport, ensuring that local performance remains predictable and unaffected by remote disruptions. Experimental results demonstrate that SM2 achieves comparable throughput to state-of-the-art alternatives such as Zenoh, while requiring significantly fewer system resources. Moreover, SM2 supports higher communication frequencies and lower latency, especially as the number of communicating ROS2 nodes increases. These results highlight SM2 as a scalable and resource-efficient middleware solution for reliable robotic applications.