The Wild

The wild is stories and ideas for AI in the wild! The phrase wild is often used within, the machine learning community, to indicate an experiment or solution that operates on data from a uncontrolled source. It's another way of saying that an AI was designed and tested on real data and not solely on a manicured experimental data set.

Somax has a tough job because it is intended for the Wild not the lab. It will leverage existing models and data sets but in the end it must be able to apply them to wild data.

School Map

This is a simple Idea for constructing a useful AI that can help students find a classrooms.

  1. Find a school (maybe yours) that can be accessed when few people are around. If you have trouble with such access, ask a science or technology teacher for assistance. When you show the teacher Somax and explain how you are applying Artificial Intelligence, you will get way, way more help than you ever needed! Be sure and ask first or be warned that your data set may be limited to the dean's office!
  2. Get a printed map of the school. The map should delineate the classrooms, doors, and halls. The map also needs to have the name of each teacher printed in their respective classroom. The map can be labeled by hand and need not be complete. However, it will not be able to find a teacher that is not listed.
  3. Start at any location in the school and execute the Mapping Decision Stream. This will load a capture engine with two forks one to collect the location data and one to collect the label data. It will also load a capture engine for known position based on data from the AHRS.
  4. Collect the data point for the location triggers and the actual location. A double tap on the rear will start data collection. In collection mode Somax will use its camera and motion controller to take pictures of the area around it. When the camera stops moving and faces you.,set your map on the floor, put the end of a finger on the map where it best indicates your current location. Pick up Somax and double tap the rear. Somax's frame display will show a portion of the map with an X indicating where it believes your are. If it is correct, double tap to use Somax's suggestion. Otherwise double tap Somax one more time and with your finger on the still the map, lean Somax towards the map until he indicates that the label has been recorded.
  5. Repeat step 4 in as many locations as possible. Use you intuition about the distance between location and where might be most helpful.
  6. Stop the Decision Stream when you feel you have enough data to get started. If you get to much, all the better. If you don't get enough you always go back an get more later.
  7. Place Somax on it's DreamBox and instruct him to learn the captured Decision Stream. The amount of data will determine how long it will need to learn the locations and mapping.
  8. Take Somax to school and test his ability to navigate. Start by loading the trained HotPlugAI that is based on your newly learned Decision Stream. Next hold somax up and double tap the back of the unit. When prompted by Somax, say or air key the teachers name. Somax will next operate the camera and motion controller to take some pictures of its current surroundings. It will take as many pictures (not stored by Somax by default) as needed to reach a level of confidence in giving advice to either move: forward, backward, right or left.
  9. Follow the camera to the target classroom. Somax will indicate the direction of travel needed to get to the target classroom using the camera as an indicator of direction.
  10. When confused, Somax will look straight up until you stop moving. When stopped Somax will again scan the environment as in step 8 to reacquire it's location. When it is acquired Somax will again point in the required direction of travel.
  11. Share Your HotPlugAI with the community so you friends can use it too!