CAN-ROS Integration: What Auto & Industry Need to Know

As someone working in automotive and industrial automation, you’ve likely faced this dilemma: How do we leverage modern robotics frameworks without compromising the real-time guarantees our systems require?

[Download the full paper here]

At CanEduDev, we’ve been deeply invested in this challenge. Our co-founder Hashem Hashem recently conducted a comprehensive study on integrating Controller Area Network (CAN) with ROS 2, now available as a preprint. Using our CanEduDev Rover platform as a testbed, we developed a modular gateway architecture that successfully bridges these two worlds – and the results reveal both exciting opportunities and important realities our industry needs to face.

The Good News 📈

Our integration achieved impressive average latencies of 170 µs and jitter of 73 µs, proving that CAN-ROS bridges can work for soft real-time applications. The modular “digital twin” approach abstracts CAN complexities while preserving critical safety features like human override capability.

Real benefits for industry:

• Enables simulation with Gazebo before physical deployment

• Unlocks powerful ROS tooling for data analysis and debugging

• Makes systems accessible to robotics developers without deep CAN expertise

The Reality Check ⚠️

Here’s what the automotive and automation community needs to know from our findings:

Performance spikes are real. While average latencies looked good, the system experienced periodic spikes reaching up to 1100 µs. For hard real-time control loops, this variability could be problematic.

Memory overhead matters. Python-based ROS nodes consumed around 80 MiB for the first node, scaling linearly with ~40 MiB per additional node – a significant footprint for embedded systems.

The bimodal latency distribution with peaks at 70-80 µs and 180-200 µs suggests underlying middleware behavior that affects timing predictability.

What This Means for Us

ROS 2 on general-purpose operating systems struggles with hard real-time guarantees – CAN remains superior for control-critical tasks. Our study confirms what many of us suspected: a hybrid approach is necessary.

The architecture works well for our Rover application, but for safety-critical applications requiring strict timing guarantees, the periodic spikes could impact system reliability.

Moving Forward

Our research suggests that rewriting in C++ could reduce memory consumption through node composition and eliminate garbage collection-related latency spikes, though at the cost of development speed and accessibility.

For those of us designing next-generation automotive and industrial systems, this research provides a crucial message: ROS 2 is a powerful tool for high-level control and development, but keep your critical loops on proven real-time protocols like CAN.

The future isn’t ROS replacing CAN – it’s intelligently combining both.

At CanEduDev, we’re committed to bridging the gap between industrial control and modern robotics frameworks. If you’re working on similar challenges or interested in discussing CAN-ROS integration for your applications, we’d be happy to share more insights and explore how our experience might benefit your projects.

Feel free to reach out – let’s advance industrial automation together! 🚀