Install ROS 2 on the Modalix SOM

This tech note describes how to install the Robot Operating System 2 (ROS 2) on the Modalix SOM DevKit and why ROS 2 is a valuable companion to SiMa.ai® silicon for physical AI workloads.

What is ROS 2?

ROS 2 is the second-generation Robot Operating System, an open-source set of software libraries and tools for building robot and physical AI applications. Rather than a monolithic operating system, ROS 2 is a middleware layer that runs on top of Linux and provides a standard way for the many independent processes inside a robot — sensors, actuators, perception models, planners, controllers — to discover one another and exchange data.

Key features include:

Node-based architecture. Each capability (a camera driver, a detector, a motor controller, a SLAM stack) runs as an independent node. Nodes are composed at runtime, so the same perception node can be reused across very different robots.
DDS-based communication. ROS 2 builds on the Data Distribution Service (DDS) standard for real-time, peer-to-peer pub/sub messaging with configurable Quality-of-Service (reliability, durability, deadline, liveliness) — a hard requirement for safety-critical and time-sensitive robotics.
Standard message types. A large catalog of common message definitions (sensor_msgs, geometry_msgs, nav_msgs, vision_msgs, etc.) lets perception, planning, and control components from different vendors interoperate without custom glue code.
Cross-platform and real-time friendly. ROS 2 runs on Linux, supports ARM64 and x86_64, and is designed to coexist with real-time kernels and microcontroller stacks (micro-ROS).
Rich tooling ecosystem. rviz2 for 3D visualization, rqt for introspection, rosbag2 for recording and replaying sensor streams, and ros2 launch for declarative system bring-up.

Why run ROS 2 on the Modalix SOM?

Physical AI — autonomous mobile robots, drones, robotic arms, AMRs, smart cameras with actuation — needs both high-throughput neural inference and a deterministic, well-understood software fabric for sensors, control, and safety. The Modalix SOM was built for exactly that combination:

Offload perception to the MLA, keep ROS 2 on the Arm cores. Perception nodes (object detection, segmentation, depth, pose, BEV fusion) can call into ModelExecutor or a GStreamer pipeline running on the MLSoC’s MLA, while ROS 2 publishers forward the results as standard vision_msgs or sensor_msgs — no proprietary IPC required.
Single board, full stack. Camera ingest, ML inference, ROS 2 middleware, planning, and motor control can all run on the same Modalix SOM, removing the latency and BOM cost of a separate companion compute board.
Power-efficient edge deployment. Modalix’s performance-per-watt envelope lets battery-powered platforms (drones, AMRs, handheld inspection units) run modern detection and tracking models in real time without thermal throttling.
eLxr Debian base. The Modalix SOM runs an eLxr Debian-derived runtime, so standard ROS 2 apt packages install cleanly — no cross-compilation or custom Yocto layers needed for the common case.

Prerequisites

Important

ROS 2 installation on the Modalix SOM requires SDK release 2.1.0 or higher running on the eLxr runtime. To check or convert the runtime, see Convert to eLxr.

You will also need:

A Modalix SOM DevKit powered on and reachable over the network or serial console.
sima-cli installed on the DevKit. If it is not already present, see sima-cli.
An active internet connection on the DevKit for package download.

Install ROS 2

sima-cli provides a one-shot installer that pulls the correct ROS 2 distribution for the Modalix SOM, configures the apt sources, and installs the base packages and standard tooling.