Pipeline Configuration

Pipelines are declared in src/config/pipeline_config.json. The WebUI Pipeline Editor reads and writes this file automatically via the /save-pipeline-config/<name> API endpoint.

Real-world example

This is a trimmed excerpt from the actual pipeline_config.json showing the key schema fields:

{
  "Test": [
    {
      "action_name": "device_input.py",
      "action_params": {
        "camera_bus_id": "basic_test",
        "frame_rotation": 0
      },
      "position": { "x": 100, "y": 100 },
      "uuid": "op-mljk5q36-arwb",
      "connections": [
        {
          "from_uuid": "op-mljk5q36-arwb",
          "from_port": "frame",
          "to_uuid": "op-mljk60f4-az78",
          "to_port": "frame",
          "data_type": "frame",
          "is_default": false,
          "custom_waypoints": null
        }
      ]
    },
    {
      "action_name": "temporal_acceleration_preprocessor_rust.py",
      "action_params": {
        "camera_bus_id": "basic_test",
        "apriltag_map_path": "{project_root}/files/apriltag_map_path/frc2025r2.json",
        "padding_factor": 0.35,
        "max_regions": 20,
        "min_region_size_px": 16
      },
      "position": { "x": 380, "y": 100 },
      "uuid": "op-mljk60f4-az78",
      "connections": [...]
    },
    {
      "action_name": "pnp_camera_localization.py",
      "action_params": {
        "camera_bus_id": "basic_test",
        "apriltag_map_path": "{project_root}/files/apriltag_map_path/frc2025r2.json",
        "jump_threshold": 2
      },
      "position": { "x": 1140, "y": 100 },
      "uuid": "op-mljk64nf-hrl6",
      "connections": [
        {
          "from_uuid": "op-mljk64nf-hrl6",
          "from_port": "camera_pose",
          "to_uuid": "op-mljk60f4-az78",
          "to_port": "camera_pose",
          "data_type": "camera_pose",
          "is_default": true,
          "custom_waypoints": [...]
        }
      ]
    }
  ]
}

Schema reference

Operation node

Field	Type	Required	Description
action_name	string	yes	Module file name including .py
action_params	object	yes	Constructor kwargs (excludes injected params)
position	{x, y}	yes	Canvas layout position (cosmetic)
uuid	string	yes	Unique node ID (op-<random>)
connections	array	yes	Outgoing edges from this node

Connection (edge)

Field	Type	Description
from_uuid	string	UUID of the source node
from_port	string	Named output port on the source
to_uuid	string	UUID of the destination node
to_port	string	Named input port on the destination
data_type	string	Informational type label (used by the editor UI)
is_default	bool	true = temporal edge (previous frame's output)
custom_waypoints	array|null	Editor canvas routing waypoints (cosmetic)

The {project_root} token

Any string in action_params containing {project_root} is substituted with the absolute path to the project root at pipeline build time. This keeps configs portable across machines:

"apriltag_map_path": "{project_root}/files/apriltag_map_path/frc2025r2.json"

Adding a new pipeline

1. Add a new top-level key to pipeline_config.json with at least a device_input.py node configured with the correct camera_bus_id.
2. Connect downstream operations as edges in the connections array.
3. Restart the backend (or click Restart in the Settings tab).

The easier route is to use the Pipeline Editor in the WebUI, which generates the JSON automatically and saves it on every change.

Validation checklist

• Every action_name exists as a file in src/main_operations/definitions/ or src/secondary_operations/.
• Every device_id in action_params matches an exact startup registry ID (cpu, cuda:N, or mx3:N).
• The connection graph is acyclic (except for temporal is_default edges, which are allowed feedback loops).
• All uuid values are unique within the pipeline.
• Paths using {project_root} resolve to files that exist on the target machine.