Polymath Robotics Blog | Polymath Perspectives from ROSCON2025 in Singapore
Polymath Perspectives from ROSCON2025 in Singapore
November 13, 2025
Troy Gibb and Emerson Knapp, two of Polymath’s Staff Engineers, attended ROSCon 2025 in Singapore a few weeks ago.
Here are their takeaways & hot takes from the conference - as jotted down by the gentlemen themselves!
Themes & Takeaways
Build and Infrastructure
- Was on full display as one of the pieces a) talked about in official presentations and b) had a lot of discussions (including a “Birds of a Feather” group discussion session coordinated by Emerson)
- Hottest items:
- nix
- Troy’s hot take: So many good chats on this one! Although we didn’t run into many folks using NixOS in their prod fleets.
- Might be worth reaching out to Yuyuan Yuan at ZettaScale, who was an avid evangelist
- Emerson notes:
- nix-ros-overlay - uses superflore to generate Nix definitions for ros distros
- ros2nix - generate Nix definitions from your ROS package.xml
- nix-ros-base - likely very outdated
- pixi
- Pixi is being presented as one of the major package managers for the ros ecosystem going forward.
- It is now the default package manager for the windows build of ros2, and if we had to guess, will start eating its way into linux distros at some point soon.
- Troy’s hot take: Pixi on the whole seems like a great way to isolate runtimes without containers and has some lightweight conventions on top of standard ros (pixi.toml), however for full OS configuration, I’m still personally more of a fan of NixOS.
- Built on Conda
- ROS integration is via robostack
- balena
- Troy had a great conversation with Guillaume Doisy, VP at Dexory, who waxed eloquent about belana usage on their fleet of 100+ robots. Incremental upgrades was a feature he called out (even though it’s container based). Otherwise didn’t hear much about it at the conference.
- balena documentation
- mender
- Emerson: I’ve been interested in this for a long time. Not mutually exclusive with any of the package management stuff, provides an A/B rootfs delta update mechanism that might replace/simplify use of CM tools like Ansible/Chef.
- Someone said they were using it just for the fleet management portion right now, which makes me think that the feature doesn’t 100% suck, worth looking at more.
- nix
Testing
- RTest (RTest GitHub) is a unit test stub for running deterministic ros2 unit tests in C++. It works by mocking out the underlying pub/sub libraries in ros2
- Troy’s hot take: worth experimenting with. The folks at Spyrosoft developed this with a client, so it is battle tested.
- replay_testing
- Completely floored the entire conference. People loved it so much they walked out early.
- Had a couple of people come up to chat more about it, and stated their intent to use it.
- RTest (RTest GitHub) is a unit test stub for running deterministic ros2 unit tests in C++. It works by mocking out the underlying pub/sub libraries in ros2
Simulation
- Some folks from Nvidia, Gazebo and Robotec.ai presented a new simulation standard: simulation interfaces
- This standard is meant to reduce friction between using different simulators— currently they target Isaac sim, Gazebo and O3DE.
- Troy’s Hot Take: Definitely worth investigating this framework more! If we have a common layer between these frameworks, the migration cost will be minimal.
- “Evaluation of ROS 2 Simulators 2025”
- The speaker here presented some benchmarks on the common simulators. TL;DR: Isaac Sim performed extremely poorly in all benchmarks.
- Gazebo
- Didn’t see anything worth writing home about
- Some folks from Nvidia, Gazebo and Robotec.ai presented a new simulation standard: simulation interfaces
Data
- Roboto.ai is building really cool tooling to run analyses on robotics datasets (e.g. bunch o bags).
Tools
- Cloudini seems like a good package to trial run. Promises up to 50% compression of point clouds, and has an extension in Foxglove for its custom data type
- Emerson’s Hot Take: The core algorithm that Davide’s research led to is actually really simple! (that’s great) But, I discussed this with him and he fully indexed on covering the “general pointcloud” use case.
- ros2_fmt_logger is a promising take at a more ergonomic logging API for ROS 2
- yazi looks like really nice terminal-based filesystem explorer
- viser is a visualizer someone recommended
- Roboflow
- @Troy Gibb got a verbal promise to get a free trial of this tool!
- Cloudini seems like a good package to trial run. Promises up to 50% compression of point clouds, and has an extension in Foxglove for its custom data type
RMW - Zenoh
- TBD - Troy missed all these talks for the testing track!
- It’s looking really promising, performance seems great and that’s all we really care about.
Our Favorite Talks
- Day 1
- Hosted BoF session, didn’t do much else other than networking
- Day 2
- ROS Project Update
- On Use of Nav2 Route Server
- Highlighted video of Polymath doing tomato harvesting in production!
- Replay Testing
- Rtest - Reliable ROS 2 Unit Tests Framework
- seems promising, we have a lot of flaky tests ourselves, the core patterns encourage it
- Zenoh Strikes Back
- Day 3
- Infrastructure Project Update
- The ROS infrastructure is the unsung hero of the whole project.
- Our Ten Most Common ROS Questions Answered
- Super informative high level talk for anybody who uses ROS 2
- Solving ROS Package Management With Pixi and Robostack
- Escape Velocity: Smarter, Cleaner ROS 2 Launch Patterns
- Evaluation of ROS 2 Simulators 2025
- Spoiler: Isaac did the worst by far
- Introducing the New ROS Simulation Standard
- Promising but early stage, maybe not much we can actually use today
- Introducing Rclrs: The ROS 2 Client Library for Rust
- good to see the language landscape expand
- Infrastructure Project Update
Polymath’s Sessions
Emerson - coordinated group discussion “Deployment & Launch Tooling”
This was a 1 hour “Birds of a feather” (BoF) session, which is just an unstructured group discussion on a topic of shared interest. This one had 30-40 people, but turned into more of an impromptu panel with 5-6 most vocal participants, Emerson moderating, with others chiming in occasionally with questions or thoughts.
High level takeaway: We still have a long way to go building the bridges between DevOps and robotics as robotics matures into an industry.
Topics that came up:
- Applications start on robot?
- SystemD services running launchfiles or individual nodes
- Containers & some form of composition, feeling disillusioned with containers for on-robot use
- What's the payload you deploy to your robot?
- discussed incremental updates because payload sizes are really big and you feel this more when trying to scale
- Fleet management (overloaded term)
- consensus is there is no clear winner solution for robots still but many tools exist to solve parts of the problem.
Troy - “Replay Testing”
ROSCON2025 - "Replay Testing" - Troy Gibb - YouTube
Watch on YouTube
Emerson - Lightning Talk “Monobranch strategy for ROS package maintenance”
SLIDES HERE
ROSCON2025 - "Lightning Monobranch" - Emerson Knapp - YouTube
Watch on YouTube
Emerson - “Escape Velocity: How to write better launchfiles”
SLIDES HERE
ROSCON 2025 - "Escape Velocity" - Emerson Knapp - YouTube
Watch on YouTube
Photos!
Emerson with his fellow members of the ROS 2 Project Management Committee
How many badge ribbons can you reasonably collect?
Troy’s groundbreaking Testing Pyramid framework
Bouldering meetup after the first day of the conference
Troy and Emerson on a Jurassic adventure in the Cloud Forest dome at the Gardens by the Bay
Robots sighted in the wild
Written By
Emerson Knapp
Emerson is a Staff Engineer at Polymath Robotics.