Add public manifest and dataset docs

DATA_DICTIONARY.md

@@ -0,0 +1,107 @@
+# Data Dictionary
+
+## Files
+
+| Path | Purpose |
+|---|---|
+| `README.md` | Dataset card and quickstart. |
+| `manifest.csv` | Human-readable experiment index. |
+| `manifest.parquet` | Parquet copy of the experiment index. |
+| `data/<config>/<public_experiment_id>.parquet` | Long-format time-series rows for one experiment. |
+| `PARQUET_VALIDATION.md` | Validation summary for the published Parquet files. |
+
+## Time-Series Parquet Schema
+
+Every experiment Parquet file has the same five columns:
+
+| Column | Type | Description |
+|---|---|---|
+| `time_ns` | `int64` | Timestamp in nanoseconds from the recorded time base. |
+| `time_utc` | `timestamp[ns, tz=UTC]` | UTC timestamp. |
+| `experiment` | `string` | Experiment identifier stored with the rows. |
+| `field` | `string` | Signal name, such as `elbow_q` or `right_knee/actual_q`. |
+| `value` | `double` | Numeric measurement value. |
+
+The data is long format: each timestamp and signal appears as a row. To work in a conventional table shape, pivot by `field`.
+
+## Manifest Columns
+
+| Column | Description |
+|---|---|
+| `public_experiment_id` | Public experiment ID used for filenames and examples. |
+| `hf_config` | Hugging Face config and subfolder containing the experiment Parquet. |
+| `hf_relative_path` | Relative path to the experiment Parquet inside the dataset repo. |
+| `robot` | Robot family, such as `h1` or `g1`. |
+| `platform` | Collection platform or source category. |
+| `experiment_group` | High-level group, such as `single_joint`, `multijoint`, or `single_joint_sysid`. |
+| `joint_group` | Joint grouping, such as arm, leg, or elbow teststand context. |
+| `joint` | Specific joint or multi-joint grouping. |
+| `motion_type` | Motion/program type parsed from the experiment name. |
+| `pose` | Pose label when encoded in the experiment name. |
+| `trial` | Trial number when encoded in the experiment name. |
+| `position` | Position label for teststand/sysid runs when available. |
+| `frequency_hz` | Frequency value parsed from the experiment name when available. |
+| `amplitude` | Amplitude value parsed from the experiment name when available. |
+| `recorded_at_from_name_utc` | UTC timestamp parsed from the experiment name when available. |
+| `min_time_utc` | Earliest timestamp in the experiment Parquet. |
+| `max_time_utc` | Latest timestamp in the experiment Parquet. |
+| `row_count` | Number of rows in the experiment Parquet. |
+| `field_count` | Number of distinct signal fields in the experiment Parquet. |
+| `parquet_size_bytes` | Size of the experiment Parquet file in bytes. |
+| `has_current` | Whether the experiment includes a current signal. |
+| `has_voltage` | Whether the experiment includes a voltage signal. |
+| `field_names` | Comma-separated list of signal fields in the experiment Parquet. |
+| `release_status` | Release readiness label, such as `ready` or `partial_electrical`. |
+| `quality_notes` | Data-quality notes for the experiment. |
+
+## Configs
+
+| Config | Files | Rows | Notes |
+|---|---:|---:|---|
+| `h1_single_joint` | 79 | 38,992,000 | H1 actual robot single-joint experiments. |
+| `h1_multijoint` | 98 | 35,856,000 | H1 actual robot multijoint experiments. |
+| `h1_elbow_teststand` | 156 | 27,660,000 | H1 elbow teststand experiments. |
+| `g1_leg_single_joint` | 126 | 74,339,640 | G1 leg single-joint experiments. |
+| `g1_leg_multijoint` | 15 | 9,353,880 | G1 leg multijoint experiments. |
+| `g1_arm_single_joint` | 84 | 34,414,560 | G1 arm single-joint experiments. |
+| `g1_arm_multijoint` | 15 | 6,062,832 | G1 arm multijoint experiments. |
+
+## Field Naming Notes
+
+H1 actual robot files use fields such as:
+
+```text
+elbow_q, elbow_dq, elbow_tau, elbow_temp
+shoulder_pitch_q, shoulder_pitch_dq, shoulder_pitch_tau, shoulder_pitch_temp
+shoulder_raise_q, shoulder_raise_dq, shoulder_raise_tau, shoulder_raise_temp
+shoulder_yaw_q, shoulder_yaw_dq, shoulder_yaw_tau, shoulder_yaw_temp
+```
+
+H1 elbow teststand files use:
+
+```text
+elbow_q, elbow_dq, elbow_tau, elbow_temp
+elbow_current, elbow_voltage
+```
+
+Seven H1 elbow teststand files are marked `partial_electrical` and include only:
+
+```text
+elbow_q, elbow_dq, elbow_tau, elbow_temp
+```
+
+G1 arm and leg files use slash-separated joint/signal names:
+
+```text
+right_elbow/actual_q
+right_elbow/command_q
+right_elbow/dq
+right_elbow/ddq
+right_elbow/position_error
+right_elbow/tau_est
+right_elbow/temperature_0
+right_elbow/temperature_1
+right_elbow/voltage
+```
+
+The same signal suffix pattern is used across the relevant G1 joints in each arm or leg config.

PARQUET_VALIDATION.md

@@ -0,0 +1,30 @@
+# Parquet Validation Summary
+
+Validated locally before upload on 2026-06-05 PDT.
+
+## Result
+
+- Manifest rows: 573
+- Parquet files under `data/`: 573
+- Files opened and read: 573
+- Missing files: 0
+- Extra files: 0
+- Empty files: 0
+- Total rows in manifest: 226,678,912
+- Total rows read from Parquet: 226,678,912
+- Total Parquet bytes in manifest: 1,857,492,979
+- Null cells in required columns: 0
+- NaN measurement values: 0
+- Infinite measurement values: 0
+- Schema mismatches: 0
+- Row-count mismatches: 0
+- Field-name mismatches: 0
+
+Overall result: pass.
+
+## Release Status
+
+- `ready`: 566 files
+- `partial_electrical`: 7 files
+
+The `partial_electrical` files are H1 elbow teststand experiments that have position/velocity/torque/temperature fields but do not include current and voltage. They are marked explicitly in the manifest.

README.md

@@ -3,25 +3,168 @@ license: apache-2.0
 tags:
 - robotics
 - sim-to-real
+- system-identification
 - h1
 - g1
 - unitree
-- RL
 - isaaclab
 - newton
+configs:
+- config_name: manifest
+  data_files: manifest.parquet
+  default: true
+- config_name: h1_single_joint
+  data_files: data/h1_single_joint/*.parquet
+  default: false
+- config_name: h1_multijoint
+  data_files: data/h1_multijoint/*.parquet
+  default: false
+- config_name: h1_elbow_teststand
+  data_files: data/h1_elbow_teststand/*.parquet
+  default: false
+- config_name: g1_leg_single_joint
+  data_files: data/g1_leg_single_joint/*.parquet
+  default: false
+- config_name: g1_leg_multijoint
+  data_files: data/g1_leg_multijoint/*.parquet
+  default: false
+- config_name: g1_arm_single_joint
+  data_files: data/g1_arm_single_joint/*.parquet
+  default: false
+- config_name: g1_arm_multijoint
+  data_files: data/g1_arm_multijoint/*.parquet
+  default: false
 ---
 
-These datasets were captured from Anchor Lab, a sim-to-real transfer laboratory from Nvidia Robotics.
+# Anchor Lab Robot System Identification Dataset
 
-This dataset is used to calibrate simulation against physical measurements for zero-shot sim-to-real deployment.
+These datasets were captured from Anchor Lab, a sim-to-real transfer laboratory from NVIDIA Robotics.
 
+This dataset is used to calibrate simulation against physical measurements for zero-shot sim-to-real deployment. It contains H1 and G1 robot system-identification time-series experiments in Parquet format.
 
-Contributors:
-* Elena Shrestha <eshrestha@nvidia.com>
-* Vignesh Bhavananthan <vbhavanantha@nvidia.com>
-* Vidur Vij <vidurv@nvidia.com>
-* Alex Omar <aomar@nvidia.com>
-* Andrew Wrenn <awrenn@nvidia.com>
-* Justin Shrake <jshrake@nvidia.com>
-* Michael Clive <mclive@nvidia.com>
-* Hong Wang <hongw@nvidia.com>
+## Start Here
+
+Start with the default `manifest` config. It has one row per experiment and tells you which Parquet file to open.
+
+Each experiment Parquet file uses long-format time-series rows:
+
+```text
+time_ns, time_utc, experiment, field, value
+```
+
+The experiment files are stored under:
+
+```text
+data/<config>/<public_experiment_id>.parquet
+```
+
+## Quickstart
+
+```python
+from huggingface_hub import hf_hub_download
+import pandas as pd
+
+repo_id = "nvidia/Anchor-Lab"
+
+manifest_path = hf_hub_download(
+    repo_id=repo_id,
+    repo_type="dataset",
+    filename="manifest.csv",
+)
+manifest = pd.read_csv(manifest_path)
+
+# Example: H1 elbow teststand experiments with current and voltage.
+subset = manifest[
+    (manifest["hf_config"] == "h1_elbow_teststand")
+    & (manifest["has_current"])
+    & (manifest["has_voltage"])
+    & (manifest["release_status"] == "ready")
+]
+
+row = subset.iloc[0]
+parquet_path = hf_hub_download(
+    repo_id=repo_id,
+    repo_type="dataset",
+    filename=row["hf_relative_path"],
+)
+
+df = pd.read_parquet(parquet_path)
+print(row["public_experiment_id"])
+print(df.head())
+```
+
+Plot one signal:
+
+```python
+import matplotlib.pyplot as plt
+
+signal = df[df["field"] == "elbow_q"].sort_values("time_ns")
+plt.plot(signal["time_utc"], signal["value"])
+plt.xlabel("time")
+plt.ylabel("elbow_q")
+plt.show()
+```
+
+Convert one long-format experiment to wide format:
+
+```python
+wide = (
+    df.pivot_table(
+        index=["time_ns", "time_utc", "experiment"],
+        columns="field",
+        values="value",
+        aggfunc="first",
+    )
+    .reset_index()
+)
+```
+
+## Configs
+
+| Config | Experiments | Description |
+|---|---:|---|
+| `manifest` | 573 | One row per experiment. Use this first. |
+| `h1_single_joint` | 79 | H1 actual robot single-joint experiments. |
+| `h1_multijoint` | 98 | H1 actual robot multijoint experiments. |
+| `h1_elbow_teststand` | 156 | H1 elbow teststand experiments. |
+| `g1_leg_single_joint` | 126 | G1 leg single-joint experiments. |
+| `g1_leg_multijoint` | 15 | G1 leg multijoint experiments. |
+| `g1_arm_single_joint` | 84 | G1 arm single-joint experiments. |
+| `g1_arm_multijoint` | 15 | G1 arm multijoint experiments. |
+
+## Common Filters
+
+```python
+# All G1 leg single-joint experiments.
+manifest[(manifest["robot"] == "g1") & (manifest["hf_config"] == "g1_leg_single_joint")]
+
+# H1 teststand experiments with electrical measurements.
+manifest[
+    (manifest["hf_config"] == "h1_elbow_teststand")
+    & (manifest["has_current"])
+    & (manifest["has_voltage"])
+]
+
+# Experiments marked ready.
+manifest[manifest["release_status"] == "ready"]
+```
+
+## Data Notes
+
+- 573 Parquet files are included.
+- All files use the same long-format schema: `time_ns`, `time_utc`, `experiment`, `field`, `value`.
+- 566 experiments are marked `ready`.
+- 7 H1 elbow teststand experiments are marked `partial_electrical`; they contain valid mechanical fields but do not include current/voltage.
+- See `DATA_DICTIONARY.md` for manifest column descriptions and signal naming notes.
+- See `PARQUET_VALIDATION.md` for the local validation summary.
+
+## Contributors
+
+* Elena Shrestha `<eshrestha@nvidia.com>`
+* Vignesh Bhavananthan `<vbhavanantha@nvidia.com>`
+* Vidur Vij `<vidurv@nvidia.com>`
+* Alex Omar `<aomar@nvidia.com>`
+* Andrew Wrenn `<awrenn@nvidia.com>`
+* Justin Shrake `<jshrake@nvidia.com>`
+* Michael Clive `<mclive@nvidia.com>`
+* Hong Wang `<hongw@nvidia.com>`

manifest.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8098251ab9b2e17788f9adc70226a1fecb659684008e437974fa62fb92f1fcc6
+size 56808