Cookie CodersFTC Team 39092
Cookie Coders // Research

Ask better questions.
Show the evidence.

Research notes from our Reinforcement Learning Environment (RLE): what we tried, what the model did, and what still needs a physical test.

Simulation, not a robot record. These studies describe uncalibrated modeled performance. Best observed results are not proven maximums or guarantees.

Research notes

3 studies // BIOBUZZ // Newest first

Full-match study6 min read

How many HIVE TIPs are possible?

Our full-match RLE study reached 13 simulated HIVE TIPs per alliance. Here is the evidence, the distribution, and why that is not a proven maximum.

Read the study
13best observed simulated TIPs
Two-robot AUTO study6 min read

Four HIVE TIPs in 30 seconds

A two-robot scripted planner completed four simulated AUTO HIVE TIPs. A paired holdout reveals the difference between a best-case witness and repeatable performance.

Read the study
4completed simulated TIPs before 30 seconds
Solo AUTO study6 min read

Three HIVE TIPs. One autonomous robot.

The solo AUTO RLE study reached three simulated TIPs with an inactive teammate. We separate the rare training frontier from frozen-policy performance.

Read the study
3best observed solo simulated AUTO TIPs
BloxBuzz and the RLE

Where to play BIOBUZZ

BloxBuzz is where you can play BIOBUZZ; the RLE is where our agents train and are evaluated. The RLE does not run in Roblox, and our goal is for it to use the same game engine, rules and physics as BloxBuzz. These results come from the RLE, not from BloxBuzz gameplay, and this study does not verify that the two match.

Opens Roblox in a new tab. Roblox's age requirements and privacy practices apply. BloxBuzz is unofficial and is not affiliated with or endorsed by FIRST or RTX.

Play BloxBuzz on Roblox

Cookie Coders / Next coordinates

The frontier keeps moving.

Research. Real machines. New minds.
Choose your next experiment.

  1. Full-match study

    How many HIVE TIPs are possible?

    Our full-match RLE study reached 13 simulated HIVE TIPs per alliance. Here is the evidence, the distribution, and why that is not a proven maximum.

    Read the study
  2. Two-robot AUTO study

    Four HIVE TIPs in 30 seconds

    A two-robot scripted planner completed four simulated AUTO HIVE TIPs. A paired holdout reveals the difference between a best-case witness and repeatable performance.

    Read the study
  3. Solo AUTO study

    Three HIVE TIPs. One autonomous robot.

    The solo AUTO RLE study reached three simulated TIPs with an inactive teammate. We separate the rare training frontier from frozen-policy performance.

    Read the study
  4. Physical frontier

    Our Robot

    From a modeled idea to a physical machine. Explore the engineering behind our robot.

    Explore the robot
  5. Learning frontier

    Chipling Academy

    Build the next generation of problem solvers. Start with curiosity, then turn it into code.

    Enter the academy
  6. Intelligence frontier

    Agent Chip

    Meet the intelligence side of Cookie Coders: a new way to explore, ask questions and learn.

    Meet Agent Chip
  7. Evidence, not hype

    Inside the evidence

    Read the methods, retained event witnesses and limitations. Simulation is a starting point, not a robot record.

    Inspect the sources
  8. BIOBUZZ / FIRST

    Know the game

    Connect the research to BIOBUZZ. Use official FIRST resources and the current rules before a physical test.

    Open game resources

Scroll or swipe to explore. Every card is a direct link.

Research studies use a Reinforcement Learning Environment (RLE). Modeled outcomes need physical validation.