> ## Documentation Index
> Fetch the complete documentation index at: https://dadd.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# useObservation

> Build policy-ready observation vectors from live MuJoCo state

Build flat observation vectors for controllers, telemetry, and RL policies without hard-coding offsets.

## Signature

```tsx theme={null}
const obs = useObservation({
  qpos: true,
  qvel: true,
  ctrl: true,
  sensors: ["imu_gyro", "imu_accel"],
  sites: ["tcp"],
  projectedGravity: "torso",
});

obs.read(): {
  values: Float32Array | Float64Array;
  layout: { name: string; start: number; size: number }[];
}
```

## Usage

```tsx theme={null}
import { useObservation, usePolicy } from "mujoco-react";

function PolicyDriver({ policy }) {
  const obs = useObservation({
    qpos: true,
    qvel: true,
    sensors: ["imu_gyro", "imu_accel"],
    projectedGravity: "torso",
  });

  usePolicy({
    frequency: 50,
    onObservation: () => obs.readValues(),
    infer: ({ observation }) => policy.predict(observation),
    onAction: ({ action, model, data }) => {
      for (let i = 0; i < Math.min(model.nu, action.length); i++) {
        data.ctrl[i] = action[i];
      }
    },
  });

  return null;
}
```

## Config

| Field              | Type                     | Description                                                        |
| ------------------ | ------------------------ | ------------------------------------------------------------------ |
| `time`             | `boolean`                | Include scalar simulation time                                     |
| `qpos`             | `boolean`                | Include all joint positions                                        |
| `qvel`             | `boolean`                | Include all joint velocities                                       |
| `ctrl`             | `boolean`                | Include all actuator controls                                      |
| `act`              | `boolean`                | Include all actuator activation values                             |
| `sensordata`       | `boolean`                | Include all raw sensor data                                        |
| `sensors`          | `string[]`               | Include named sensor values in order                               |
| `sites`            | `string[]`               | Include named site world positions in order                        |
| `projectedGravity` | `string \| string[]`     | Include world gravity projected into each named body's local frame |
| `output`           | `"float32" \| "float64"` | Output vector type. Defaults to `Float32Array`                     |

## `buildObservation`

Use the pure helper when you already have `model` and `data` inside a physics callback:

```tsx theme={null}
import { buildObservation, useBeforePhysicsStep } from "mujoco-react";

useBeforePhysicsStep(({ model, data }) => {
  const obs = buildObservation(model, data, {
    qpos: true,
    qvel: true,
    projectedGravity: "torso",
  });

  console.log(obs.values, obs.layout);
});
```

Missing named resources are skipped. Use the returned `layout` as the source of truth for the vector generated from the current model.