SteadySync-Equilibria

Overview

This project implements visual servoing for a Kinova Gen3 7-DOF robotic arm in ROS 1 Noetic. The robot uses image-based visual servoing (IBVS) to detect and track a rectangular object using its end-effector camera. The system extracts corner points of the rectangle and adjusts the robot's position dynamically to align with a predefined reference.

Features

• Real-time Rectangle Detection: Uses ORB feature detection and RANSAC filtering for robust corner point detection.
• ROS Integration: Subscribes to camera image topics, processes frames in OpenCV, and publishes detected features.
• Trajectory Control: Uses FollowJointTrajectoryAction to move the Kinova Gen3 arm.
• Visual Servoing Loop: Computes position corrections based on detected rectangle corner deviations.
• Gazebo Simulation Support: Can run in both real hardware and Ignition Gazebo.

System Architecture

+------------------------------------------------------+
|                      ROS Framework                   |
+------------------------------------------------------+
|   Image Capture    |   Feature Detection   | Control |
|  (Camera Topic)    | (ORB + RANSAC)       | (MoveIt!)|
+------------------------------------------------------+
|                   Kinova Gen3 Robot Arm             |
+------------------------------------------------------+

Dependencies

Ensure you have the following installed:

• ROS 1 Noetic
• MoveIt!
• OpenCV (cv_bridge, image_transport)
• Kinova ROS packages (kortex_driver, kortex_examples)
• Gazebo Fortress/Garden (for simulation)

Installation

# Clone your workspace and install dependencies
cd ~/Desktop/robot_ws/src
git clone https://github.com/your-repo/your-project.git
cd ..
rosdep install --from-paths src --ignore-src -r -y
catkin_make
source devel/setup.bash

Running the Project

1. Launch Gazebo Simulation

roslaunch kortex_examples gazebo3.launch

2. Start the Visual Servoing Node

rosrun kortex_examples visual_servoing.py

3. View Processed Images

rqt_image_view /edge_detected_image

4. Move Robot to Initial Position

rostopic pub /move_robot std_msgs/Empty {}

How it Works

1. Capturing Reference Image: The arm moves to position-1 and captures a reference image.
2. Real-time Processing: The camera continuously captures frames and detects the largest rectangle.
3. Feature Extraction: ORB detects corner points, filtered using RANSAC for robustness.
4. Error Computation: Compares current rectangle corners with the reference and computes an error vector.
5. Robot Adjustment: Adjusts the arm’s position iteratively to align with the reference.
6. Convergence: The process stops once the alignment error falls below a threshold.

Expected Results

• The robot should accurately align with the rectangle even with slight perturbations.
• Robust corner detection should work even in varying lighting conditions.

Future Improvements

• Refine ORB & RANSAC parameters for better robustness.
• Implement deep learning-based corner detection for improved accuracy.
• Extend to 6-DOF pose estimation instead of just 2D alignment.

Authors

• Your Name – Surabhi Dwivedi, Deepraj Majumdar
• Institution/Organization – Indian Institute of Technology, Jodhpur

---

⭐ If you find this project useful, give it a star! ⭐