ROS 2 ROS 2 fault diagnosis
A missing topic message becomes a diagnostic casebook. Compare rules and optional Jev Choice on fabricated events while preserving unknown and review outcomes.
OPEN RESEARCH / OPEN COLLABORATION
Robotics experiments worth investigating.
Explore ROS 2, Kubernetes, cloud–edge AI, agentic systems and Jev through animated conceptual traces, rigorous research questions and contributor-ready experiments.
Start here
This is a set of proposed research scenarios on ROS 2, Kubernetes and cloud-edge AI robotics; none has a measured result yet.
It is for students, teachers and researchers who want to study these questions. The synthetic starting points need no robot, GPU or paid account.
- Pick a scenario from the cards below.
- Step through its animated walkthrough with the Play walkthrough button on the scenario page.
- Open the linked GitHub issue to collaborate on that scenario.
OPEN RESEARCH PROGRAMSIMULATION FIRST
ARCHITECTURE ILLUSTRATION
01 / Choose your track
FLAGSHIP / SCENARIO 01
FLAGSHIP / SCENARIO 02
FLAGSHIP / SCENARIO 03
FLAGSHIP / SCENARIO 04 FUTURE / SCENARIO 05
FUTURE / SCENARIO 06
Start with a problem that matters.
These pages are crawlable, technical and honest about their status. An animated walkthrough explains a **proposed** data flow; it is not a model demonstration or evidence of a working robot fleet.
6 scenarios shown
Edge vs cloud inference placement
Explore which inference location is eligible when latency budgets, cloud disconnects, model resources and action freshness matter.
Agent skill screening
Trace how an LLM's high-level robot skill proposal is checked in code before any optional Jev second opinion.
Multi-robot task assignment
Study a fictional scout, carrier and arm; rank only eligible candidates while deterministic scheduling retains authority.
Robot observability incident routing
Investigate whether Choice and Score improve the classification and human routing of synthetic ROS 2/Kubernetes alerts beyond labelled monitoring rules.
Propose a scoped subtaskHugging Face VLA compatibility review
Ask whether Jev adds useful advisory context after deterministic checks of robot embodiment, action space, camera geometry, license and hardware.
Review prerequisite #57No matching scenario. Try another topic.
02 / Experiments, not anecdotes
STEP 01 / PLAN
STEP 02 / MEASURE
STEP 03 / SHARE
From a question to a reproducible result.
Pre-register the comparison
Define the null hypothesis, task, fixture split, deterministic baseline, model revision, metrics, confidence limits and failure conditions before testing.
Collect full evidence
Record raw predictions, abstentions, timeouts, run manifests, seeds and licenses. Keep model calls opt-in and never run a paid provider in default CI.
Publish methods and limitations
Require raw data, analysis scripts, independent review and explicit limitations before showing measured outcomes. Negative and inconclusive results count.
Reality check: this repository's original Vagrant/Foxy/Kubernetes 1.29 environment is historical. The Jazzy smoke test demonstrates only image-local communication. Cross-worker DDS, Jev runtime integration and robotics task-success metrics are still unverified. This website is a research invitation, not a safety-certified robotics product.
03 / Result transparency
No invented speedups or accuracy charts.
What will eventually appear?
Per-scenario experiment IDs, a clearly labelled status, comparable baseline/JeV runs, measurable outcomes, uncertainty, reproducibility manifests and links to raw evidence.
Today every listed experiment is planned and every performance value is N/A. Those are intentional, not omitted success stories.
Open evidence registry
See the status of all four flagship research tracks and the requirements for publishing verified data. Contribute a reproducible result or a negative finding through an issue/PR.
View experiments & results ↗
04 / For students, teachers and collaborators
Choose the contribution that fits your experience.
BSC / FIRST CONTRIBUTION
Synthetic data & tutorials
Improve a ROS 2 casebook, catch ambiguous labels, repair an example or evaluate readability. No hardware or paid account required.
Start with #84MSC / REPRODUCIBLE EXPERIMENT
Implementation & methods
Build a deterministic baseline, a simulator contract or a network fault matrix; publish tests and explicit run conditions.
Explore #85PHD / OPEN RESEARCH
Novel comparisons & evaluation
Preregister a hypothesis on calibration, placement regret, agent review or scheduling fairness and analyze honest evidence.
Explore #88Want to collaborate with Claude Code or Codex?
Every scenario issue includes a bounded task prompt, dependencies, validation expectations and safety boundaries. Read AGENTS.md before editing. AI coding assistance is optional and does not replace human review.
Follow the epic & open issues ↗ Help audit SEO & accessibility ↗
05 / Frequently asked questions
Before you join a study.
Do I need a robot, GPU or paid account?
No for the synthetic casebooks, design exercises and offline rules. Actual accelerated inference, live Jev and any robot/simulator benchmarks require separately documented equipment, licenses and permission.
Does clicking Play call Jev?
No. The walkthrough uses preauthored educational text and client-side highlighting, never hosted inference. Jev's role is optional research only. It is not a motor controller or permission oracle.
How do we submit results?
Start with the experiment registry and evidence issue #88. Include a reproducible manifest, raw outcomes, code and negative cases. Unverified claims will not appear as measured results.
Can this become a thesis or class project?
Yes, the scenarios are designed as research prompts and teaching starting points, but feasibility, novelty, ethics approval, licensing and academic supervision must be established for each proposed project.