Step-by-Step Tutorials

The tutorials follow one robot data pipeline: acquire SensorFrame, produce a ParsedFrame, create a Plan, then send a ControlCommand to a control loop. Each chapter introduces one business problem and only the APIs needed to solve it.

Start with your first task and submitting functions and data, then proceed through:

1. Prioritize control commands: urgent control and ordinary analysis share a pool.
2. Delayed retry and health checks: retry later and stop observable soft-periodic work.
3. Batch sensor frames: choose a batch path based on whether each task needs a future.
4. Load, sense, then plan: run planning only after prerequisite work completes.
5. Bounded waiting and status: finish safely before a phase change or shutdown.
6. Fan out events with Topic: give independent consumers separate bounded queues, backpressure policies, and shutdown ownership.
7. Complete robot pipeline: connect startup dependencies, frame streams, configuration, commands, snapshots, diagnostics, and shutdown.
8. Service data import: apply the same dependency, batch, partial-failure, and bounded-drain model to a server request.

Every page identifies scale and concurrency assumptions, ownership of asynchronous objects, failure injection, in-flight work during exit, and when changing requirements require a different abstraction. Treat those sections as an architecture-review checklist.

For API selection by problem, see Choose a Submission API and Execution Models and Routing Boundaries. Enter real-time, GPU, lock-free, or Blocking I/O topics only when their constraints are explicit.