jdvakil/prox_test

Proximity Learning Depth Sensor Dataset

Franka Panda robot with multi-link depth sensors in Isaac Lab warehouse simulation for proximity-based robot learning.

Dataset Description

This dataset contains depth sensor readings from 39 distributed sensors mounted across the Franka Panda robot's links, collected in Isaac Lab simulation. The data is designed for training diffusion policies for collision-aware manipulation.

Dataset Summary

Property	Value
Robot	Franka Panda (7 DOF)
Episodes	1
Frames	180
FPS	60
Sensors	39 depth sensors
Sensor Resolution	8x8 pixels each
Simulation	Isaac Lab (Warehouse)
Format	Parquet (Snappy compression)

Data Fields

• frame_index (int64): Frame number in the episode
• timestamp (float64): Simulation timestamp in seconds
• action (float32[7]): Joint position commands for 7 DOF arm
• proprioception (float32[14]): Joint positions (7) + joint velocities (7)
• observations (dict): Dictionary containing depth readings from 39 sensors, each with shape [8, 8]

Sensor Configuration

The 39 depth sensors are distributed across the robot's links to provide egocentric proximity sensing for collision avoidance during manipulation tasks.

Usage

Loading with Pandas

import pandas as pd

df = pd.read_parquet("hf://datasets/jdvakil/prox_test/data.parquet")

# Iterate through frames
for idx, row in df.iterrows():
    frame_idx = row['frame_index']
    timestamp = row['timestamp']
    action = row['action']  # shape: [7]
    proprioception = row['proprioception']  # shape: [14]
    observations = row['observations']  # dict of sensor readings
    
    # Access individual sensor
    for sensor_name, sensor_data in observations.items():
        depth = sensor_data['depth']  # shape: [8, 8]
        print(f"{sensor_name}: {depth.shape}")

Loading with Datasets Library

from datasets import load_dataset

dataset = load_dataset("jdvakil/prox_test")
print(dataset)

# Access a sample
sample = dataset['train'][0]
print(sample.keys())

Intended Use

This dataset is intended for:

• Training diffusion policies for proximity-aware manipulation
• Research on egocentric depth sensing for robotics
• Collision avoidance using distributed sensor arrays
• Sim-to-real transfer of proximity-based control

License

MIT License