Overview

4-member team project demonstrating autonomous robot navigation using only camera-based perception without LiDAR, deployed on TurtleBot4 hardware.

Core Capabilities

Lane Following

• ArUco Markers: Visual lane definition and position tracking
• Perspective Geometry: Calculate robot position relative to markers
• Smooth Control: Proportional heading adjustment for lane centering

Stop Sign Detection

• YOLOv8n Model: Real-time inference (30+ FPS on embedded hardware)
• Confidence Filtering: Robust multi-frame validation
• 3D Localization: Projective geometry for sign position

Obstacle Detection

• Depth Estimation: Monocular depth perception
• Safety Layer: Collision avoidance while pursuing tasks
• Real-time: Integrated into control loop

Technical Stack

• Platform: TurtleBot4 with integrated RGB camera
• Vision: OpenCV, YOLOv8 (nano), ArUco detection
• Framework: ROS 2
• Processing: Embedded GPU (Jetson Nano)

Getting Started

# Clone project
git clone https://github.com/svaibhav101/perception-nav.git
 
# Install dependencies
rosdep install --from-paths src -y
 
# Build and run
colcon build && ros2 launch perception_nav navigation.launch.py

Validation

• ✅ Gazebo simulation testing
• ✅ Real TurtleBot4 hardware deployment
• ✅ Lane following accuracy: 95%+
• ✅ Stop sign detection: 98% confidence
• ✅ Obstacle avoidance: Collision-free navigation