Behavior Control

Team Lead: Daeyon Kim and Nikolas Margaritis

Team Description: Develops the intelligent behavior that enables our robots to coordinate with each other and make the best possible choices for what skills to use at each moment of the game.

Multi-agent Behavior

  1. Multi-agent RL
  2. Dynamic positioning
  3. Passing behavior

Obstacle Avoidance

  1. Teammate avoidance
  2. Opponent avoidance

Game logging (Collect real trajectory)

Record what the robot saw and decided during a game, then replay it offline to debug and test behavior changes without the robot

  1. Record game data with ros2 bag
  2. Compare behavior before and after code changes

Learning-based Behavior

We can collect data in sim or use real data from Game logging

  1. RL-based skill selection
  2. End to end behavior policy
  3. Behavior tree replacement

Head Behavior

  1. Active head control
  2. Ball tracking
  3. Landmark scanning
  4. Ball search