PitVit/1000_Robot_Manipulation_Tasks

1000 Robot Manipulation Tasks

This dataset is currently undergoing a phased upload process. Due to its substantial size, the complete data is being incrementally published to ensure stability and integrity during transmission. We appreciate your patience as we finalize the full availability of the dataset.

This dataset is distributed under the MIT License. This permissive free software license grants users the freedom to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software (or data, in this context). It also includes a disclaimer of warranty, meaning the data is provided "as is" without guarantees of fitness for a particular purpose.

---

Dataset Overview

This dataset comprises a comprehensive collection of autonomous robot rollouts, encompassing both successful task completions and observed failures. These rollouts were gathered from a large-scale experimental evaluation designed to assess robotic manipulation capabilities.

The experiment involved the execution and evaluation of 1000 distinct manipulation tasks utilizing a diverse set of over 400 unique objects. While this dataset represents the majority of the rollouts collected during the experiment, it does not include every single recorded instance. Nevertheless, it offers a robust and representative sample for analysis and downstream research.

---

Data Structure and Contents

The dataset is organized within a top-level folder named rollouts. Inside this folder, you will find numerous subfolders, each corresponding to a unique robot rollout. The naming convention for these rollout folders is Method_TaskName_RolloutNumber.

For example, a folder named MT3_close_black_metal_bottle_0000 signifies the very first rollout (0000) performed for the task of "closing the black metal bottle" by the MT3 method.

Each individual rollout folder contains the following files:

• end_effector_poses.npy: A NumPy array storing the 3D poses (position and orientation) of the robot's end-effector over time.
• end_effector_velocities.npy: A NumPy array containing the translational and rotational velocities of the end-effector over time.
• gripper_is_open.npy: A binary NumPy array indicating whether the robot's gripper was open (1) or closed (0) at each timestamp.
• head_camera_depth_images.npy: A NumPy array stacking depth images captured by the robot's head-mounted camera.
• head_camera_rgb_images.npy: A NumPy array stacking RGB (color) images captured by the robot's head-mounted camera.
• joint_angles.npy: A NumPy array of the robot's joint angles at each time step.
• joint_velocities.npy: A NumPy array of the robot's joint velocities at each time step.
• success.txt: A plain text file containing either the string "True" or "False", indicating whether the corresponding rollout was successful or a failure, respectively.
• task_name.txt: A plain text file containing a string representing the name of the specific task performed in this rollout (e.g., "close_black_metal_bottle").
• skill_name.txt: [DEPRECATED] This file is a repetition of the task name and is maintained for historical compatibility but is not recommended for current use.

---

Accessing the Data

To facilitate easy access and integration into your research, each rollout can be loaded as a Python dictionary. Below is a snippet demonstrating how to load the data from a single rollout folder:

import numpy as np
import os

def load_rollout(rollout_path: str) -> dict:
    """
    Loads all data files for a given rollout into a dictionary.

    Args:
        rollout_path (str): The file path to the specific rollout folder.

    Returns:
        dict: A dictionary where keys correspond to the data attribute names
              and values are the loaded data (NumPy arrays or strings).
    """
    rollout_data = {}

    # Define the files and their corresponding dictionary keys
    files_to_load = {
        "end_effector_poses.npy": "end_effector_poses",
        "end_effector_velocities.npy": "end_effector_velocities",
        "gripper_is_open.npy": "gripper_is_open",
        "head_camera_depth_images.npy": "head_camera_depth_images",
        "head_camera_rgb_images.npy": "head_camera_rgb_images",
        "joint_angles.npy": "joint_angles",
        "joint_velocities.npy": "joint_velocities",
        "success.txt": "success",
        "task_name.txt": "task_name",
        "skill_name.txt": "skill_name", # Deprecated, but included for completeness
    }

    for filename, key in files_to_load.items():
        file_path = os.path.join(rollout_path, filename)
        if not os.path.exists(file_path):
            print(f"Warning: File not found at {file_path}. Skipping.")
            continue

        if filename.endswith(".npy"):
            rollout_data[key] = np.load(file_path)
        elif filename.endswith(".txt"):
            with open(file_path, 'r') as f:
                rollout_data[key] = f.read().strip() # .strip() to remove leading/trailing whitespace
        else:
            print(f"Warning: Unknown file type for {filename}. Skipping.")

    return rollout_data

# Example usage:
# Assuming 'rollouts' is the main directory containing all rollout folders
# and 'MT3_close_black_metal_bottle_0000' is a specific rollout folder.
try:
    # Replace 'path/to/your/dataset/rollouts/MT3_close_black_metal_bottle_0000'
    # with the actual path to a rollout folder on your system.
    example_rollout_folder = "path/to/your/dataset/rollouts/MT3_close_black_metal_bottle_0000"
    
    # Create dummy files for demonstration if the path doesn't exist
    if not os.path.exists(example_rollout_folder):
        print(f"Creating dummy rollout data at {example_rollout_folder} for demonstration.")
        os.makedirs(example_rollout_folder, exist_ok=True)
        np.save(os.path.join(example_rollout_folder, "end_effector_poses.npy"), np.random.rand(10, 6))
        np.save(os.path.join(example_rollout_folder, "end_effector_velocities.npy"), np.random.rand(10, 6))
        np.save(os.path.join(example_rollout_folder, "gripper_is_open.npy"), np.random.randint(0, 2, 10))
        np.save(os.path.join(example_rollout_folder, "head_camera_depth_images.npy"), np.random.rand(10, 64, 64))
        np.save(os.path.join(example_rollout_folder, "head_camera_rgb_images.npy"), np.random.randint(0, 256, (10, 64, 64, 3)))
        np.save(os.path.join(example_rollout_folder, "joint_angles.npy"), np.random.rand(10, 7))
        np.save(os.path.join(example_rollout_folder, "joint_velocities.npy"), np.random.rand(10, 7))
        with open(os.path.join(example_rollout_folder, "success.txt"), "w") as f: f.write("True")
        with open(os.path.join(example_rollout_folder, "task_name.txt"), "w") as f: f.write("close_black_metal_bottle")
        with open(os.path.join(example_rollout_folder, "skill_name.txt"), "w") as f: f.write("close_black_metal_bottle")


    loaded_rollout = load_rollout(example_rollout_folder)

    # You can now access different components of the rollout
    print("\nSuccessfully loaded rollout data:")
    for key, value in loaded_rollout.items():
        if isinstance(value, np.ndarray):
            print(f"- {key}: NumPy array of shape {value.shape}")
        else:
            print(f"- {key}: {value}")

except Exception as e:
    print(f"\nAn error occurred during example usage: {e}")
    print("Please ensure you replace 'path/to/your/dataset/rollouts/MT3_close_black_metal_bottle_0000'")
    print("with a valid path to a rollout folder from your downloaded dataset.")