Grasp Simulation and Data Collection

---

This project demonstrates a simulation of robotic grippers using PyBullet. The simulation enables the gripper to perform random grasp attempts on an object (e.g., a small cube), evaluate the grasp quality, and save the collected data for further analysis.

---

Features

1. Random Grasping:
   The gripper performs randomized grasps using noise-based position generation around the object.

2. Grasp Quality Evaluation:
   Determines whether the grasp is successful by evaluating the grasping duration without slippage.

3. Data Collection:
   Saves gripper and object positions, orientations, and grasp quality labels (success/failure) to a CSV file.

4. GUI and Headless Modes:
   Supports running the simulation with GUI visualization or in headless mode for batch data collection.

   • GUI Mode: Used for debugging or observing the simulation process.

   • Headless Mode: Used for efficiently running simulations in batches, saving resources and time.

---

Dependencies

• Python 3.6+
• Download env_requirement.txt.
  Run the following command to install all the packages:

  pip install -r requirements.txt

---

Usage

Function Parameters

GripperType - Select the type of gripper to grasp object,PR2 / F3

where - Select the direction around the object to generate random grasp

ObjectName - Select the object to grasp.

batch_size - Number of grasps to simulate and save the data in data.csv

Mode - 0 for Debugging mode, 1 for DataColletion

---

Run Single Grasp in GUI

For debugging or visualizing the grasping process, use:

a=RUN('F3',**1**,'cube',500,0)

This will load the environment, perform a single grasp at selected direction (1), and display the simulation in a GUI.

Batch Data Collection

Without GUI (Headless Mode)

For faster batch data collection:

a=RUN('F3',1,'cube',**100**,**1**)

• Simulates 100 grasps at selected direction without rendering the GUI.
• Saves the data in data.csv.

With GUI

To visualize each grasp while collecting data, Change the parameters of the following code from DIRECT to GUI.
This code is under DataCollection(self) method in classes.

p.connect(p.GUI)

---

Data Structure

The collected data is saved in a CSV file (data.csv) with the following fields:

Column	Description
Ori_x, Ori_y, Ori_z, Ori_w	Orientation of the gripper in quaternion format.
Pos_x, Pos_y, Pos_z	Position of the gripper (base frame).
Obj_Pos_x, Obj_Pos_y, Obj_Pos_z	Position of the object.
Label	Grasp quality: 1 for success, 0 for failure.

---

Simulation Workflow

1. Initialization:
   Sets up the environment with a table, a small cube, and a PR2/F3 gripper.

2. Grasp Execution:

   • Generates a random position around the object.
   • Moves the gripper to the target position and closes its fingers to grasp the object.
   • Lifts the object to evaluate the grasp.

3. Data Saving:
   Captures the positions, orientations, and grasp quality in a structured format and appends them to the CSV file.

---

Key Functions

Function	Description
random_position	Generates a random grasp position around the object.
update_gripper_direction	Moves the gripper to the target position and orientation.
grasp	Closes the gripper's fingers to grasp the object.
lifting	Lifts the object slowly after grasping.
getdata	Captures and logs positions and orientations of the gripper and object.
is_successful	Evaluates grasp quality based on contact points.
SaveData	Saves collected data to a CSV file.

---

Environment Setup

• Objects:

  • PR2/F3 Gripper (pr2_gripper.urdf)
  • Table (table.urdf)
  • Cube (cube_small.urdf with global scaling).

• Physics:

  • Gravity: -10 m/s².
  • Simulation Steps: Continuous during grasping and lifting.

---

Example Output

A snapshot of the collected data in data.csv:

Ori_x,Ori_y,Ori_z,Ori_w,Pos_x,Pos_y,Pos_z,Obj_Pos_x,Obj_Pos_y,Obj_Pos_z,Label
0.0,1.0,0.0,0.0,0.05,0.0,0.5,0.0,-0.4,0.62,1
-0.2,0.98,0.0,0.0,0.04,-0.1,0.6,0.02,-0.38,0.65,0
...

---

Future Improvements

• Add support for different object shapes and sizes.
• Introduce reinforcement learning for better grasping strategies.

---

License

This project is licensed under the MIT License.