Autonomous Mobile Robot for Color-Based Object Retrieval
Designed and developed a mobile robot capable of autonomously detecting, navigating to, and retrieving colored cubes in a controlled environment. This project combined computer vision, object localization, and motion control to execute reliable pick-and-place operations based on color recognition.
Highlights:
- Camera Calibration: Mapped 2D image coordinates to 3D spatial positions using intrinsic and extrinsic camera properties.
- Vision-Based Detection: Identified object color and shape with high precision under varying conditions.
- Real-World Localization: Determined both object and robot positioning within a known reference frame.
- PID-Controlled Motion: Smooth and accurate robot movement using feedback-based trajectory correction.
Technical Workflow:
1. Camera Calibration
Goal: Establish a relationship between image pixels and real-world coordinates.
- Intrinsic Calibration: Obtained using 30 checkerboard image sets in MATLAB to derive focal length, principal point, and lens distortion.
- Extrinsic Calibration: Used known corner positions of the arena to determine the camera’s position and orientation relative to the environment.
2. Color and Shape Recognition
- Captured frames processed in real time.
- Converted color space from BGR to HSV for more effective segmentation.
- Applied color masks to isolate target hues (blue, green, magenta).
- Calculated centroids using image moments.
- Differentiated cubes from patches based on contour area.
- Transformed detected centroid coordinates to world space.
3. Object and Robot Localization
Using the derived calibration parameters, both the robot’s and object’s global positions were calculated.
- Depth Calculation: Depth (
Z_w) was computed first, followed by lateral (X_w,Y_w) world coordinates.
4. Autonomous Navigation
Control Strategy:
- A PID controller regulated wheel velocities to direct the robot toward the target.
- Navigation relied on minimizing orientation error between the robot and the goal point.
Tuning Details:
- Initially modeled in Simulink for parameter estimation:
Kp= 1.317,Ki= 0.5039,Kd= -0.1388
- Final values adjusted manually based on field performance:
Kp= 1,Ki= 0,Kd= 0
Real-World Applications:
- Warehouse Automation: Efficient retrieval and sorting systems.
- Manufacturing: Automated material handling solutions.
- Research Platforms: Prototyping intelligent robotic behaviors.