The Story Behind the Game:
Today many companies, such as IBM, Google, OpenAI, Nvidia and others, are actively developing artificial intelligence programs. BEST Robotics does not want to be left behind in this exciting endeavor and is looking to its many teams to develop and train an AI system. Teams will compete to see which is the BEST.
Your team will build a robot which represents a pair of AI software systems. The first is a neural network constructor. Neural nets are the building blocks of AI, defining the connections from the inputs of the AI to the outputs of the AI. Of course, better AIs will have more complex connections within their network. Use your robot to select the connection matrix you will build and then wire it up.
The second is a “web crawler,” which searches the internet to find data that will be used to train the AI. But BEST does not want to use just any data. There is an old computer term from 1957 (although the concept goes back at least to Charles Babbage) that states “garbage in, garbage out.” In other words, if you put flawed, biased, or poor quality data into a computer program you will get similarly poor quality answers out of the program. BEST only wants to train their AI using the highest quality information available. As you might imagine, some sources of information (libraries and newspapers) contain more truth and fewer errors than other sources (social media sites). While you could trust luck, it might be better to have a robot that can distinguish between fact and fiction...
Your robot’s twofold job is to build a neural network and to “crawl” the internet sites available collecting the TRUE factoids and ignoring the FALSE factoids.
The Rules -- a summary
The field represents an AI system with 4 distinct color-coded networks and several sources of data for training. Team owned areas and components are color-coded. Each network consists of a Neural Network, Network Selector Box, and a Factoid Sorting Tray. Each team manages their own Neural Network within the system for a specific purpose. There are Network Selectors that are used to establish which of three AI models a team may create. The field is arranged in identical quadrants. However, there are no boundaries and robots are free to maneuver anywhere on the field.
Tasks/Objectives
Design and build a robot capable of training your team's Artificial Intelligence (AI) by performing the following tasks:
- Data Collection
- Collect data elements from any of 4 distinct areas
- Sort data elements as TRUE or FALSE based on certain (physical) properties
- Select/Build an AI Model
- Select a specific AI model to build/train (one of 3)
- Build a Neural Network AI model by connecting network Tiles to the available nodes
- Train (LEARN)
- Feed the model as much TRUE data as possible to maximize the probability of success for the model
The tasks are to be performed by robotic methods.
Scoring for the AI system tasks are given in the table that follows:
| Task Completed | End of Match Location/Condition | Points Each | Max # Possible Per Team |
| Library of Content Golden Factoid Collected | Resting on the teams' robot, inside the Sorting Tray, or inside a Neural Network node | 100 | 1 |
| Library of Content Factoid (white) Collected | Resting on the team's robot, inside the Sorting Tray, or inside a Neural Network node | 100 | 1 |
| Any Factoid Collected | Resting on the team's robot | 10 | 40 + 20 + 16 + 2 |
| Any Factoid Sorted | Inside the Sorting Tray | 5 | 40 + 20 + 16 + 2 |
| Any Factoid Installed | Inside a Neural Network node | 15 | 13 * 9 = 117 |
| WSD Gate Latch Released | Latch is released when it is past vertical, or it no longer prevents the gate from falling. Subsequent latch releases are worth fewer points. | 100, 30, 10, 4 | 4 |
| Tweeter Tower Activation | Activated when the diamond plunger is depressed, and the factoids are released. | 7 | 4 |
| 2-Node Neural Net Connections | Lines on the connection Tile are touching two Neural Network nodes (Painted Zone) | 20 | 4 |
| 3-Node Neural Net Connections | Lines on two Connection Tiles touch one common node and two additional Neural Network nodes (Painted Zone) | 60 | 2 |
| AI Model Selected | Neural Net Selector selects any model (i.e., not Blank) | 20 | 4 |
| Bonuses | |||
| Live Net Connection Bonus | A 2-Node Neural Net Connection with at least one net TRUE at each end of the connection | 30 | 4 |
| Live Net Connection Bonus | A 3-Node Neural Net Connection with at least one net TRUE at each of the 3 connection points | 90 | 2 |
| Network Configuration Match |
The Neural Network Configuration matches the model selected by the Neural Net Selector |
100, 125, 150 | 1 |
Field Overview
Factoids!
| Game Piece | Description | Image |
| TRUE Failbook Factoid |
1.6" (41mm) diameter plastic golf balls |
 |
| FALSE Failbook Factoid |
1.6" (41mm) diameter plastic golf balls |
|
| TRUE WSD Factoid |
1 1/2" PVC pipe with caps/inserts, 4" long, white |
 |
| FALSE WSD Factoid |
1 1/2" PVC pipe with caps/inserts, 4" long, white |
|
Wall Street Gate
There are 20 Factoids in the Wall Street Dispatch, of which 16 are TRUE. Factoids are evenly distributed throughout the WSD at the beginning of the match. Factoids in the Wall Street Dispatch are 1.66" diameter x 4.5" and cylindrical in shape.
Tweeter Towers
Library of Content
| Game Piece | Description | Image |
| Library of Content Golden Factoid |
3" PVC pipe, |
 |
| Library of Content White Factoid | 3" PVC pipe,
3.5" diameter x 3" high, |
 |
Factoid Sorting Tray
This tray is used to store Factoids deposited by the robot in the Robot Starting Area (RSA) and manually sorted by the Spotter.
- Robot places Factoids into the RSA.
- Spotter sorts the Factoids into the appropriate TRUE or FALSE compartment of the Sorting Tray.
- Spotter can only retrieve the Factoids when the robot is completely outside the RSA.
- Factoids touched by the spotter must be placed in the Sorting Tray; they cannot be left in the RSA, and they cannot be removed from the Sorting Tray once sorted.
Neural Network
Neural Net Model Selector
The handle is horizontal (Blank) by default. There are 3 models that can be selected, clockwise from the Spotter's view:
- Model A (Generative)
- Model B (Machine Learning)
- Model C (Computer Vision)
The Spotter can view the selection from the viewport on the top of the selector and confirm the robot has picked the desired model.
SA BEST Factoids Field
Here is the final setup of the 2025 San Antonio BEST Factoids field and pit area.
