Software Development System

Targeting Fields:

OPENCV and SLAM

Finding and locating the ball, field lines, robots and goal net.

Currently able to detect all objects except robots

Robot Navigation and Reinforcement Learning

Reverse Engineering from 2D/3D AI champion code from previous years for development of our own AI code, written in Python. AWS Deep Cloud.

Communication Systems with ROS and Data Collection

Being able to get good connectivity and make a full scalable deployed application.

 

Challenges

Real-time object recognition

Two approaches:

  • OpenCV for simple features
    • Playing field
    • Lines
  • Deep Learning for object recognition
    • Ball
    • Other robots

Difficulties:

  • Limited computational power
    • Optimization
    • Asynchronous processing
    • Better hardware? GPU -> FPGA