Datasets:
metadata
license: cc-by-4.0
task_categories:
- robotics
tags:
- LeRobot
- animacy
- expressive-robots
- vla
- lamp
configs:
- config_name: default
data_files: data/*/*.parquet
squaredcuber/animacy-lamp-lerobot
Human conversational motion retargeted to animacy/lamp as a LeRobot v3.0 dataset, produced by
animacy (scripts/export_lerobot.py). The robot joint
vector is observation.state, the next frame is action, and every frame carries the
speech features, the speaking flag and the canonical human channels it was computed from,
so the same dataset trains robot-space and human-space policies.
53 episodes, 26774 frames at 30 fps (14.9 min), 5 task strings.
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("squaredcuber/animacy-lamp-lerobot")
item = ds[0] # observation.state, action, observation.human, observation.audio_features, observation.speaking, ...
Features
| key | dtype | shape | names |
|---|---|---|---|
observation.state |
float32 | [5] | base_yaw, base_pitch, elbow_pitch, wrist_roll, wrist_pitch |
action |
float32 | [5] | base_yaw, base_pitch, elbow_pitch, wrist_roll, wrist_pitch |
observation.human |
float32 | [14] | head_yaw, head_pitch, head_roll, head_x, head_y, head_z, brow_l, brow_r, ... |
observation.audio_features |
float32 | [66] | mel_00, mel_01, mel_02, mel_03, mel_04, mel_05, mel_06, mel_07, ... |
observation.speaking |
float32 | [1] | |
observation.environment_state |
float32 | [67] | mel_00, mel_01, mel_02, mel_03, mel_04, mel_05, mel_06, mel_07, ... |
timestamp |
float32 | [1] | |
frame_index |
int64 | [1] | |
episode_index |
int64 | [1] | |
index |
int64 | [1] | |
task_index |
int64 | [1] |
Episodes
One episode per contiguous face-tracked run (>= 3 s) of a source clip, split into pieces of
<= 20 s. action[t] = observation.state[t+1] inside a run (the run's last frame holds).
Joint values are in the robot's own units (see the names and the robot's ROBOT.md).
observation.environment_state duplicates the speech features + speaking flag under the key
lerobot's ACT / Diffusion policies require when there is no camera.
Sources and licenses
| clip | episodes | frames | license | source |
|---|---|---|---|---|
| cbp_vlog_day2 | 2 | 783 | Public Domain | Video Blog- CBP Super Bowl LII Countdown to Kickoff - Day 2.webm |
| kende_interview_2014 | 12 | 6112 | CC-BY-3.0 | Internet Hall of Fame 2014 Michael Kende interview.webm |
| obama_2014_09_13 | 12 | 6708 | Public Domain | 2014-09-13 President Obama's Weekly Address.webm |
| obama_2015_02_07 | 9 | 5341 | Public Domain | 2015-02-07 President Obama's Weekly Address.webm |
| royal_society_cloke | 16 | 7253 | CC-BY-3.0 | Interview on extreme weather with physical geographer Hannah Cloke – The Royal Society.webm |
| sd_rapper_interview | 2 | 577 | Public Domain | Local San Diego Rapper Interview |
Source videos are public-domain or CC-BY works (license evidence is recorded per clip in the animacy human-motion dataset); the derived motion is released CC-BY-4.0 with attribution to the sources above. No no-derivatives or non-commercial material is included by construction.
meta/info.json
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
5
],
"names": [
"base_yaw",
"base_pitch",
"elbow_pitch",
"wrist_roll",
"wrist_pitch"
]
},
"action": {
"dtype": "float32",
"shape": [
5
],
"names": [
"base_yaw",
"base_pitch",
"elbow_pitch",
"wrist_roll",
"wrist_pitch"
]
},
"observation.human": {
"dtype": "float32",
"shape": [
14
],
"names": [
"head_yaw",
"head_pitch",
"head_roll",
"head_x",
"head_y",
"head_z",
"brow_l",
"brow_r",
"brow_furrow",
"torso_lean_fwd",
"torso_lean_side",
"torso_yaw",
"mouth_open",
"smile"
]
},
"observation.audio_features": {
"dtype": "float32",
"shape": [
66
],
"names": [
"mel_00",
"mel_01",
"mel_02",
"mel_03",
"mel_04",
"mel_05",
"mel_06",
"mel_07",
"mel_08",
"mel_09",
"mel_10",
"mel_11",
"mel_12",
"mel_13",
"mel_14",
"mel_15",
"mel_16",
"mel_17",
"mel_18",
"mel_19",
"mel_20",
"mel_21",
"mel_22",
"mel_23",
"mel_24",
"mel_25",
"mel_26",
"mel_27",
"mel_28",
"mel_29",
"mel_30",
"mel_31",
"mel_32",
"mel_33",
"mel_34",
"mel_35",
"mel_36",
"mel_37",
"mel_38",
"mel_39",
"mel_40",
"mel_41",
"mel_42",
"mel_43",
"mel_44",
"mel_45",
"mel_46",
"mel_47",
"mel_48",
"mel_49",
"mel_50",
"mel_51",
"mel_52",
"mel_53",
"mel_54",
"mel_55",
"mel_56",
"mel_57",
"mel_58",
"mel_59",
"mel_60",
"mel_61",
"mel_62",
"mel_63",
"log_energy",
"delta_log_energy"
]
},
"observation.speaking": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"observation.environment_state": {
"dtype": "float32",
"shape": [
67
],
"names": [
"mel_00",
"mel_01",
"mel_02",
"mel_03",
"mel_04",
"mel_05",
"mel_06",
"mel_07",
"mel_08",
"mel_09",
"mel_10",
"mel_11",
"mel_12",
"mel_13",
"mel_14",
"mel_15",
"mel_16",
"mel_17",
"mel_18",
"mel_19",
"mel_20",
"mel_21",
"mel_22",
"mel_23",
"mel_24",
"mel_25",
"mel_26",
"mel_27",
"mel_28",
"mel_29",
"mel_30",
"mel_31",
"mel_32",
"mel_33",
"mel_34",
"mel_35",
"mel_36",
"mel_37",
"mel_38",
"mel_39",
"mel_40",
"mel_41",
"mel_42",
"mel_43",
"mel_44",
"mel_45",
"mel_46",
"mel_47",
"mel_48",
"mel_49",
"mel_50",
"mel_51",
"mel_52",
"mel_53",
"mel_54",
"mel_55",
"mel_56",
"mel_57",
"mel_58",
"mel_59",
"mel_60",
"mel_61",
"mel_62",
"mel_63",
"log_energy",
"delta_log_energy",
"speaking"
]
},
"timestamp": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"frame_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"episode_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"task_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
}
},
"total_episodes": 53,
"total_frames": 26774,
"total_tasks": 5,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": null,
"robot_type": "animacy/lamp",
"splits": {
"train": "0:53"
}
}