Bot Bowl II

See the results further down on this page.

Bot Bowl II is an AI competition in Blood Bowl using the FFAI framework. The traditional board size of 26×15 squares with 11 players on each side is used. Participants are, however, limited to use a prefixed human team. In future competitions, we plan to allow all teams and the option to customized rosters.

Bot Bowl II had six bot submissions and we added the Random agent baseline to the competition. The competition consisted of a round-robin tournament where all bots played every other bot 10 times. The winner is the one with most wins, then draws, then TDs, then CAS.

Bot Bowl II had two prizes:

  1. $500 to the 1st place winner.
  2. $500 to the highest ranked machine learning bot*.

Participants CAN win both prizes, if they win the competition with a machine learning bot.

Format

The format of Bot Bowl II is similar to last year’s competition. Only a prefixed Human team can be used. None of the pre and post-game procedures are used. Rosters are thus reset without injuries and improvement rolls.

In contrast to last year there will be no final.

Rules

Computational Resources

The philosophy of Bot Bowl is to limit the computational resources at run-time to that of a regular laptop. We of course encourage participants to use as much computation as they can afford to train/evolve their bots before the competition. Don’t expect any GPUs.

Participants

Giaobot by Zhang Pengpeng, Gao Huiru, Xu Zhiwei

Giaobot is a machine learning bot trained with IMPALA using an LSTM policy, and self-play.

Grodbot MKII by Peter Moore

Grodbot MKII is an improved version of last year’s winner, including bug fixes, less screening and more emphasis on chasing the ball.

Gotebot by Mattias Bermell

Gotebot is a machine learning bot trained using A2C, self-play, and curriculum learning. Ontop of the learned behavior, it has to simple scripted rules: Never do GFI and never end the turn with unused players left.

Minigrod by RogueLichen

Minigrod is a scripted bot based on Grodbot (the winner of Bot Bowl I) it applies dynamic formations and more focus on safe actions.

Sapling by Jonas Busk

Sapling is a search-based bot that uses Monte Carlo Tree Search with a scripted heuristic function.

Scripted Plus Bot by Mark Christiansen

Scripted Plus Bot is a scripted bot based on the one from the tutorials. It has dynamic formations, prioritizes blitzing players near the ball, and has logic for crowd surfing (pushing players of the board).

Results

  Minigrod GB MKII Scripted PB Gotebot Sapling Random Giaobot W/D/L TD CAS
Minigrod - 2/5/3 8/2/0 7/3/0 10/0/0 10/0/0 10/0/0 47/10/3 151/12 121/48
GrodBot
MKII
3/5/2 - 4/2/4 10/0/0 10/0/0 10/0/0 10/0/0 47/7/6 144/13 70/65
Scripted
Plus Bot
0/2/8 4/2/4 - 8/2/0 10/0/0 10/0/0 9/1/0 41/7/12 97/25 93/50
 Gotebot 0/3/7 0/0/10 0/2/8 - 9/1/0 7/3/0 10/0/0 26/9/25 49/45 56/61
Sapling 0/0/10 0/0/10 0/0/10 0/1/9 - 0/10/0 1/9/0 1/20/39 1/104 38/65
 Random 0/0/10 0/0/10 0/0/10 0/3/7 0/10/0 - 0/10/0 0/23/37 0/116 18/48
Giaobot 0/0/10 0/0/10 0/1/9 0/0/10 0/9/1 0/10/0 - 0/20/40 0/127 12/71

Winners

Congratulations to Minigrod (RogueLichen) for winning Bot Bowl II and to Gotebot for winning the ML prize. The authors of both bots will each recieve $500.

Highlights

Watch the highlight video here: https://www.youtube.com/watch?v=qajXQhrBuV0.