We had the pleasure of attending ROSCon 2024 in Odense, Denmark. ROSCon isn’t just a place to see what’s happening in the ROS ecosystem; it’s also a fantastic venue to explore the broader robotics industry and see what advancements have been made year over year.
This year, we noticed two major trends:
1. The fusion of AI and robotics is becoming more ingrained in robotics software and development tools.
2. The ecosystem in and around ROS continues to mature.
The Fusion of AI and Robotics
While this trend has been ongoing, it dominated the conference this year, shaping the overall tone of both sessions and after-hours conversations. The industry seems to be shifting away from the traditional kinematics-based model of robotics, focusing instead on using AI to drive the next wave of solutions in this space. One highlight was Robotec’s presentation on RAI, which uses a multi-agent approach to bring generative AI to robotics development. This project incorporates interfaces like image and speech recognition with ROS features such as topics and services. It will be exciting to watch as the Robotec team tackles some of its current limitations in the coming year.
Remedy Robotics also presented a survey on foundational models in robotics, along with a number of libraries to help get started with them. (Note: the repositories weren’t posted with this talk, but we’ll update this with the relevant projects once the video is available.)
Sony presented their work on ros2ai, a very interesting LLM-based CLI assistant for ROS 2, which aims to make ROS development more accessible and efficient.
The ROS Ecosystem Continues to Evolve
Zettascale and Intrinsic presented Zenoh, a new middleware option for ROS. Until recently, we had only seen DDS-based middleware supported by ROS; however, Zenoh breaks new ground by being purpose-built for low latency, location-transparent, unified communication at scale—and it’s not based on DDS.
Zenoh is designed for heterogeneous networks (systems where traffic spans LAN, WAN, and edge environments), unlike DDS. Nodes can dynamically discover each other and adapt to network conditions, functioning much like a peer-to-peer network, which makes it more scalable and flexible than DDS-based solutions.
Some people have noted that Zenoh can be more challenging to use out of the box than DDS, but we believe that, as it evolves and more case studies emerge, this gap will close. DDS remains the default in ROS, so moving away from it involves some extra effort.
Picknik Robotics showed off Datatamer, a library that makes exporting data to PlotJuggler easier for debugging.
Ekxide also presented on iceoryx2, a Rust-based middleware option still in early development. It promises zero-copy data transfer for ROS middleware, with Rust’s safety and concurrency features enhancing its reliability.
Finally, UC Berkeley showcased Scenic, a tool that allows teams to probabilistically generate scenarios for training set augmentation.
This was the best ROSCon yet, with record attendance and a palpable feeling that robotics is ready for “Primetime.” See you next year in Singapore!
Note: For more info on the talks at ROSCon, you can visit the official ROSCon page.
