734: Humanoid Robot Soccer — with the Dutch RoboCup Team

Vision & Sound

In the realm of humanoid robot soccer, vision and sound detection are pivotal for effective gameplay. Dario Catarrinho explains that these robots utilize convolutional neural networks for visual tasks, such as detecting the ball and field lines, which are crucial for navigation and localization on the pitch 1. Sound detection is equally important, with algorithms in place to recognize the whistle amidst ambient noise, ensuring the robots respond promptly to game cues 2. This sophisticated integration of machine learning allows the robots to operate autonomously, making real-time decisions without human intervention 2.

These robots, as I said, were fully autonomous. They make a bit of noise, so hopefully it doesn't pick up too much.

--- Dario Catarrinho

   

Reinforcement Learning

Reinforcement learning is being explored to enhance the strategic capabilities of soccer robots. Currently, the robots operate on an expert system with hard-coded behaviors, such as the nearest robot going for the ball 1. Dario Catarrinho mentions that the team is in the early stages of implementing reinforcement learning to improve decision-making, like path planning and determining which robot should pursue the ball 1. This approach aims to refine the robots' adaptability and strategic responses during matches, moving beyond simple programmed actions 3.

We are trying to get their behavior to be run on reinforcement learning. So we're talking path planning, making decisions on which robot goes for a ball.

--- Dario Catarrinho