Overview

The Objective

The goal of this project was to transform a base PenguinPi mobile robot, initially capable of only rudimentary, low level motion, into a semi/fully autonomous system. We were tasked with engineering a robot that could navigate a structured arena, map its environment, avoid hazards (represented by ArUco markers), and identify and move to specific target items (fruits and vegetables) from a predefined "shopping list."

Development

System Architecture and Milestones

Development of the system was down into several milestones:

• Kinematics & Movement: The base PenguinPi platform initially only supported rudimentary, low-level motion. We modelled the robot's kinematics to bridge this gap and calibrated it ensuring that high-level navigational decisions were accurately translated into precise, integrated motor control.

  
  

• Mapping & Localisation: The team implemented SLAM (Simultaneous Localisation and Mapping) alongside classical computer vision techniques. This enabled the robot to detect ArUco markers for accurate pose estimation while constructing a real-time map of its surroundings.

  
  

• Object Detection & Target Recognition: For the task, the system needed to identify its specific goals. A machine learning classifier was developed and integrated to successfully identify and visually differentiate the target fruit objects from other environmental noise.

  
  

• Path Planning & Obstacle Avoidance (my focus): Relying on the live SLAM map data, I implemented a modified A* path-planning algorithm that involved gaussian obstacles to allow for decision making based on weighting for safety versus distance. This architecture provided the system with global obstacle-aware navigation, allowing it to compute the most efficient routes to the target items while actively avoiding the hazards effectively.