Skip to content

dataset_stat.md

Latest commit

 

History

History
201 lines (164 loc) · 12.3 KB

dataset_stat.md

File metadata and controls

201 lines (164 loc) · 12.3 KB

📊 StyleDrive Dataset Statistics

This page provides key statistics and structure details for the StyleDrive dataset, designed for benchmarking personalized end-to-end autonomous driving.

Dataset Comparison

Dataset Reality Scenarios CL/OL E2E Style
nuScenes Real City OL ✓ ✘
OpenScene Real City & Rural OL ✓ ✘
Longest6 Sim City CL ✓ ✘
CARLA Sim City CL ✓ ✘
MetaDrive Sim City CL ✓ ✘
Bench2Drive Sim City CL ✓ ✘
NAVSIM Real City & Rural Semi-CL ✓ ✘
HuiL-RM HuiL Ramp-Merge OL ✘ ✓
HuiL-CF HuiL Car-Following OL ✘ ✓
HuiL-LC HuiL Lane-Change OL ✘ ✓
HuiL-Mul HuiL City OL ✘ ✓
UAH Real City & Highway OL ✘ ✓
Brain4Cars Real Lane-Change & Merge OL ✘ ✓
PDB Real City OL ✘ ✓
StyleDrive (Ours) Real City & Rural Semi-CL ✓ ✓

📁 Dataset Overview

Attribute Value
Total Scenarios ~29,856
➤ Training Set 25,649 scenarios
          • A (Aggressive) 3,814
          • N (Normal) 20,469
          • C (Conservative) 1,411
➤ Testing Set 4,096 scenarios
          • A (Aggressive) 538
          • N (Normal) 3,275
          • C (Conservative) 236
Scenario Types 11
Annotated Attributes 30+ motion, safety, and context fields
Style Labels Aggressive / Normal / Conservative
Source Dataset OpenScene (built upon nuPlan)
Annotation Method Rule-based heuristics + VLM-based inference + Human-in-the-loop

Dataset Collection Location (same as nuPlan, but subset)

Pittsburgh Boston Las Vegas Singapore
Pittsburgh Boston Las Vegas Singapore

🧭 Traffic Scenario Distribution

Scenario Type Train Count Test Count Total Count
Lane Following 13,385 2,345 15,730
Lane Change 316 54 370
Protected Intersection 3,670 582 4,252
Unprotected Intersection 1,191 254 1,445
Crosswalks 1,509 241 1,750
Side Ego to Main 641 60 701
Side to Main Ego 705 97 802
Countryside Road 255 97 352
Roundabout Entrance 115 13 128
Roundabout Interior 30 5 35
Carpark Areas 68 39 107
Special Interior Road 3,809 262 4,071
Total 29,856 4,096 33,952

🧭 Pie Chart Visualization

Dataset Overview Image

Dataset Detailed Structure and Access Guide

The StyleDrive dataset consists of driving scenarios, each described by a rich set of attributes related to vehicle motion, safety, contextual information, and unique scene features. Here is a detailed explanation of the data structure with data types, example data entries, and how to access this data programmatically.

🧳 Dataset Structure

Core Features

Each dataset entry contains the following core features that represent the ego vehicle's state, motion, safety, and context. The data type for each feature is specified below.

Feature Description Data Type Example
vx_ego Ego vehicle's velocity along the x-axis (m/s) List[float] [4.11, 4.14, 4.16, ...]
vy_ego Ego vehicle's velocity along the y-axis (m/s) List[float] [0.01, -0.04, -0.03, ...]
v_ego Ego vehicle's speed magnitude (m/s) List[float] [4.11, 4.14, 4.17, ...]
ax_ego Ego vehicle's acceleration along the x-axis (m/s²) List[float] [0.04, 0.08, 0.04, ...]
ay_ego Ego vehicle's acceleration along the y-axis (m/s²) List[float] [-0.65, -0.64, -1.18, ...]
a_ego Ego vehicle's overall acceleration (m/s²) List[float] [0.65, 0.65, 1.18, ...]
yaw Ego vehicle's yaw angle (radians) List[float] [2.16, 2.09, 1.98, ...]
yaw_diff Change in yaw angle between frames (radians) List[float] [-0.12, -0.22, -0.31, ...]
min_front_frame Minimum distance to the front vehicle (meters) List[float] [9, 22.69]
v_avg Average speed of ego vehicle over the last 10 frames (m/s) float 4.76
v_std Standard deviation of speed over the last 10 frames (m/s) float 0.67
vy_max Maximum lateral velocity in the last 10 frames (m/s) float 0.13
a_max Maximum acceleration in the last 10 frames (m/s²) float 2.34
ax_avg Average longitudinal acceleration over the last 10 frames (m/s²) float 0.46
a_std Standard deviation of acceleration over the last 10 frames (m/s²) float 0.59
ini_direction_judge Initial direction of the ego vehicle (e.g., "forward", "backward") string "right"
scenario_type Type of driving scenario (e.g., "lane_following", "crosswalks") string "protected_intersections"
scene_token Unique identifier for each scene string "a631fec170525388"
has_left_rear Boolean indicating presence of vehicle in the left rear zone bool false
has_right_rear Boolean indicating presence of vehicle in the right rear zone bool true
left_rear_min Minimum distance to the vehicle in the left rear zone (meters) float or null null
right_rear_min Minimum distance to the vehicle in the right rear zone (meters) float 9.69
speed_mode Speed-related preference label (Aggressive, Normal, Conservative) string "accelerating"
front_frame Temporal distances to the front vehicle across multiple frames (meters) List[float] [-1, -1, -1, -1, 22.69]
safe_frame Safety flags for each frame (indicates safe or unsafe driving condition) List[bool] [-1, -1, -1, -1, 1]
safe_ratio Proportion of frames labeled as "safe" float 1.0
unsafe_ratio Proportion of frames labeled as "unsafe" float 0.0
oversafe_ratio Proportion of frames with excessive safety margin float 0.0

🧑‍💻 Example Dataset Entry

Example Data (For Scenario: protected_intersections)

{
  "vx_ego": [4.11, 4.14, 4.17, 4.21, 4.35, 4.60, 4.95, 5.31, 5.66, 6.08],
  "vy_ego": [0.01, -0.04, -0.03, -0.04, -0.01, 0.04, 0.07, 0.13, 0.13, 0.13],
  "v_ego": [4.11, 4.14, 4.17, 4.21, 4.35, 4.60, 4.95, 5.31, 5.66, 6.08],
  "ax_ego": [0.04, 0.08, 0.04, 0.13, 0.41, 0.65, 0.80, 0.75, 0.73, 0.99],
  "ay_ego": [-0.65, -0.64, -1.18, -1.43, -1.58, -1.85, -2.17, -2.22, -1.79, -1.67],
  "a_ego": [0.65, 0.65, 1.18, 1.43, 1.63, 1.96, 2.31, 2.34, 1.94, 1.94],
  "yaw": [2.16, 2.09, 1.98, 1.83, 1.65, 1.44, 1.23, 1.00, 0.82, 0.67],
  "yaw_diff": [-0.13, -0.22, -0.31, -0.36, -0.40, -0.43, -0.45, -0.36, -0.29, -0.29],
  "min_front_frame": [9, 22.69],
  "v_avg": 4.76,
  "v_std": 0.67,
  "vy_max": 0.13,
  "a_max": 2.34,
  "ax_avg": 0.46,
  "a_std": 0.59,
  "ini_direction_judge": "right",
  "scenario_type": "protected_intersections",
  "scene_token": "a631fec170525388",
  "has_left_rear": false,
  "has_right_rear": true,
  "left_rear_min": null,
  "right_rear_min": 9.69,
  "speed_mode": "accelerating",
  "front_frame": [-1, -1, -1, -1, 22.69],
  "safe_frame": [-1, -1, -1, -1, 1],
  "safe_ratio": 1.0,
  "unsafe_ratio": 0.0,
  "oversafe_ratio": 0.0,
  "ANC_result": "N",
  "scene_scenario": "protected_intersections",
  "lane_change_frame": -1,
  "token": "a631fec170525388",
  "pedestrians": "No",
  "with_lead": "Far"
}

🧑‍💻 Example Code to Access Data

Here’s an example in Python to access the dataset attributes:

import json

# Load dataset from a JSON file (assuming the file is in your working directory)
with open('path_to_file.json', 'r') as f:
    data = json.load(f)

# Access specific scene (for example, scene with token 'a631fec170525388')
scene_data = data['a631fec170525388']

# Print velocity data of the ego vehicle
print("Ego Vehicle Velocity (vx_ego):", scene_data['vx_ego'])

# Access the scenario type
print("Scenario Type:", scene_data['scenario_type'])

# Calculate the average speed (v_ego)
v_avg = sum(scene_data['v_ego']) / len(scene_data['v_ego'])
print("Average Speed:", v_avg)

# Check if there is a lead vehicle in the scene
if scene_data['with_lead'] != 'No':
    print("Lead vehicle present:", scene_data['with_lead'])

🧑‍💻 Accessing Scene-Specific Features

For scenario-specific features like pedestrians or merge_risk, you can directly access them as:

# Example: Access pedestrians feature for a scenario
print("Pedestrians present:", scene_data.get('pedestrians', 'Unknown'))

# Example: Access merge_risk for a scenario
print("Merge risk level:", scene_data.get('merge_risk', 'None'))