NAOSAMI

All behavior and math files are in core/python/behaviors/. The data collection files are in the top-level directories beacon_data/ and beacon_data_sim/.

Notable files:

• core/python/behaviors/do_maths.py
  • runs the EM algorithm on the collected training data
• core/python/behaviors/naosami.py
  • collects the training data by randomly choosing actions and observing beacon heights
• core/python/behaviors/interpolate_actions.py
  • interpolates the action table using K-nearest neighbors and linear interpolation
• beacon_data/obs_data.txt
  • holds statistics on the collected training data such as the number of total actions executed
  • the number of unique actions (out of 40) executed
  • the number of times each beacon has been seen
  • the percentage of frames where the Nao saw all NaNs (no beacons)

Instructions on how to run our code:

1. In the tool, run the naosami behavior to collect real world data.
2. Run the format_data.sh script in the beacon_data_sim/ top-level folder to convert to a compressed numpy format.
3. Run the do_maths.py Python script to run the EM algorithm on the compressed files.
4. For the action interpolation, run the interpolate_actions.py Python script on the saved outputs of the do_maths.py script.

Miscellaneous Notes:

• For collecting real data, change the do_maths.py and format_data.sh to point to beacon_data/ the top-level folder and the files in that folder.
• If you want to collect data in simulation, run the GoalieSim behavior in the world window tool.
• If you want to teleport the player, you can press the ctrl key and then right click on the world