Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions docs/MANIFESTO.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
# ROSGraph — Direction

## Why

Robotics engineers spend too much time on ROS plumbing — writing boilerplate, debugging invisible wiring, and keeping launch files in sync with code — instead of building their application.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

as systems get larger, they become harder to reason about. The lack of a "well defined" interface (topic, service, action) contracts makes this even harder. It also excludes any ability to have automated tools to give leverage over this problem

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Addressed — incorporated into the MANIFESTO Why section alongside Emerson's comment. The paragraph now covers both the undocumented interfaces point and the scaling/automated tooling angle.


The main interfaces of ROS systems (topics, parameters, services, actions) are undocumented by default. As systems grow larger they become harder to reason about, and the lack of well-defined interface contracts blocks automated tooling from helping.

## What

A declarative, observable ROS graph. Engineers declare what their system should be; tooling generates the code and entities as needed, and verifies the running system matches the spec.

## How

1. **Language** — a formal spec to describe node interfaces and system graphs.
2. **Tooling** — translate declarations into working code.
3. **Verification** — compare spec against reality, both at runtime and statically before launch.
121 changes: 121 additions & 0 deletions docs/ROSGRAPH.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,121 @@
# rosgraph — Technical Proposal

> **Status:** Proposal
> **Date:** 2026-02-22
> **Parent:** [MANIFESTO.md](MANIFESTO.md) (direction)

---

## Executive Summary

ROS 2 has no standard schema for declaring node interfaces and no
production-ready tooling for verifying that a running system matches its
declared architecture. The ecosystem is fragmented across single-purpose
tools with overlapping scope and bus factors of one.

Key gaps — no existing tooling:

- **Graph diff** (expected vs. actual)
- **Graph linting** (pre-launch static analysis)
- **CI graph validation**
- **Node API documentation** (hand-written only today)
- **QoS static analysis** (breadcrumb is early-stage/partial)

### The Problem, Concretely

Today in ROS 2:

- Node A publishes `/cmd_vel` as `Twist`. Node B subscribes to
`/cmd_vel` as `String`. You discover this at runtime — or don't,
because the subscriber silently receives nothing.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

along this line of thought - refactoring topic names across a system is difficult (e.g. you change your teleop node to produce /cmd instead of /cmd_vel and aren't sure what the downstream implications are)

- A publisher uses `BEST_EFFORT` QoS, a subscriber uses `RELIABLE`.
DDS refuses the connection. A warning is logged but easy to miss in
a busy console. The subscriber just never gets messages.
- A node crashes mid-deployment. The rest of the system keeps running.
Nobody knows until a customer reports a failure 20 minutes later.
- You rename a parameter. Three launch files reference the old name.
`colcon build` succeeds. The system launches. The parameter silently
takes its default value.
- You rename a topic from `/cmd_vel` to `/cmd`. Several downstream
nodes subscribed to the old name silently receive nothing. There is
no static analysis to tell you what depended on it.

These are real, common bugs in production ROS 2 systems.

### Components

rosgraph is composed of the following components, ordered by priority.
These components may be wrapped by user interfaces (e.g. a CLI), but
are designed as independent, composable libraries.

1. **Node Spec (NoDL)** — a formal, machine-readable schema for
declaring node interfaces (`interface.yaml`). This is the most core
part of the project; everything else builds on it.

2. **Code Generation** — `nodl-generator` takes NoDL input and outputs
code for ROS client libraries (rclcpp, rclpy, rclrs). Must be
installable as part of a ROS distro (`apt-get install`). Requires a
plugin/sidechannel architecture so additional client libraries
(e.g. rcljava) can be supported without modifying the core generator.

3. **Runtime Discovery** — introspect a running system and produce NoDL
specs from observed nodes. Enables brownfield adoption: point at an
existing system, generate `interface.yaml` files for every node, then
iteratively refine them. Discovery and runtime monitoring (component 5)
share the same mechanism — observe the live graph, produce a spec,
diff against declared. The distinction is cadence: one-time migration
vs. continuous verification.

4. **Node-level Unit Testing** — verify a single node conforms to its
declared spec in isolation.

5. **Graph Analysis & Comparison** — integration-level verification.
Static analysis checks the full graph for type mismatches, QoS
incompatibilities, and missing connections before launch. Runtime
monitoring continuously diffs the declared graph against the live
system, flagging drift (crashed nodes, unexpected topics, QoS
changes) as it happens.

6. **Documentation Generation** — produce API documentation directly
from NoDL specs.

> **Open question:** implementation language for the generator tooling.

### Key Insight

**A declaration without code generation is a non-starter.** NoDL
proved this. The schema must generate code, documentation, and
validation to stay in sync with reality. `interface.yaml` is
simultaneously the source for code generation, the lint target for
static analysis, the contract for runtime verification, and the
reference for documentation.

### Example

A minimal `interface.yaml`:

```yaml
schema_version: "1.0"

publishers:
- topic: ~/chatter
type: std_msgs/msg/String
qos: { reliability: RELIABLE, depth: 10 }

parameters:
publish_rate:
type: double
default_value: 1.0
description: "Publishing rate in Hz"
validation:
bounds<>: [0.1, 100.0]
```

From this single file, the tooling can:
- **Generate** a typed C++/Python node context with publishers and validated parameters — no boilerplate
- **Lint** the full workspace graph for type mismatches and QoS incompatibilities before launch
- **Monitor** the running system and flag drift from the declared spec
- **Discover** a running system's interfaces and produce draft specs for brownfield adoption
- **Document** the node's API automatically

---