Configuring the Inference Pipeline

Pipeline order: observation → preprocessors → runner → postprocessors → action output. See docs/how-to/inference/configure-pre-post-processing.md.

Core code:

• src/physicalai/inference/componentfactory.py — ComponentRegistry, instantiatecomponent, MAXCOMPONENT_DEPTH.
• src/physicalai/inference/model.py — builds processor chains from manifest specs.
• Built-ins under preprocessors/ and postprocessors/; runners under runners/.

Workflow

1. Read the manifest slice for preprocessors, postprocessors, and model.runner.

- Done when: you know whether specs use type (registry short name) or class_path.

1. Prefer type for built-ins registered in component_factory (e.g. normalize/denormalize patterns in docs).
2. Use classpath + initargs for explicit classes:

``yaml preprocessors: - classpath: physicalai.inference.preprocessors.StatsNormalizer initargs: artifact: stats.safetensors ``

- Done when: initargs paths resolve relative to the export directory via resolveartifact.

1. Add a new built-in processor:

- Implement subclass of Preprocessor / Postprocessor in the appropriate package. - Register a short type name in componentfactory if manifest-friendly aliases are needed. - Add unit tests under tests/unit/inference/preprocessors/ or postprocessors/. - Done when: manifest using type or classpath instantiates in a minimal InferenceModel test.

1. Nested components in initargs must stay within MAXCOMPONENTDEPTH; avoid cyclic specs.

Validation loop

uv run pytest tests/unit/inference/preprocessors tests/unit/inference/postprocessors tests/unit/inference/test_manifest.py -q

Required checks

• Processor order matches training/export semantics (normalization before runner, denormalization after).
• Artifact file names in manifests do not traverse paths (.., absolute paths).
• New public processors appear in docs/reference/inference-api.md or how-to docs when user-visible.
• Runner choice (SinglePass, chunking runners) stays consistent with predictactionchunk vs select_action docs.

References

• docs/reference/manifest-schema.md
• docs/how-to/inference/use-manifest.md