Projects

Autonomous Object Identification

Project Overview

This project focused on programming a TurtleBot to autonomously navigate a mapped environment, locate objects, and identify image tags placed on them. Key features of the project included:

  • Robot Operating System (ROS) for navigation and control.
  • Kinect 360 Sensor for image recognition and depth sensing.
  • AMCL for localization and move_base for navigation.

Task Requirements

The TurtleBot was required to:

  • Complete the circuit within 5 minutes.
  • Return to the starting location with object identification results.

Key Strategies

Navigation

  • Nearest-Neighbor Algorithm:
    • Optimized object visitation order by minimizing Euclidean distances between objects.
  • Object Positioning:
    • Calculated valid positions relative to objects to ensure clear detection.

Angled-Approach

Object Identification

  • OpenCV's SURF Algorithm:
    • Utilized for feature detection and matching, with FLANN for template matching.
  • Fine-Tuned Parameters:
    • Addressed discrepancies between simulated and real-world conditions.
  • Confidence Levels:
    • Determined matches for template images, blank images, and duplicates based on confidence thresholds.

Feature-Detection

Robot Design and Implementation

Sensory Systems

  • Depth Sensor: Detected obstacles and aided navigation.
  • RGB Camera: Captured images for object identification.
  • Bumper Sensors: Detected collisions and complemented depth sensors for blind spots.
  • Odometry: Provided relative position and orientation data to ensure accurate localization.

Controller Architecture

  • High-Level Control:
    • Determined the robot's global path and object interaction strategy.
  • Low-Level Control:
    • Managed manual movement in high-cost map areas and handled obstacle avoidance in real-time.

Visualization of high-cost region on cost map

High-Cost-Map

Performance Metrics

  • Successfully completed the circuit with reliable object detection in real-world tests.
  • Navigation and object identification were effective but highlighted areas for improvement:
    • Path Planning: Transition from Euclidean metrics to map-based distances for accuracy.
    • Advanced Image Recognition: Integrate machine learning models such as YOLO or CNNs.

Recommendations for Improvement

  1. Path Planning:
    • Implement frontier-based exploration for dynamic path planning.
  2. Image Recognition:
    • Adopt machine learning techniques for enhanced accuracy.
  3. Testing:
    • Conduct extensive trials in diverse environments to ensure robust performance.
Return to Portfolio