Bot Bowl III

Bot Bowl III is the third AI competition in Blood Bowl and is set to happen (virtually) at the IEEE Conference of Games 2021. Submit your bot by July 15 to participate in the competition and have chance at winning one of the prestigious prizes.


Bot Bowl III will have the same format as last year while the framework offers a few new features that will empower bot developers.

Each bot will play 10 matches of Blood Bowl against each other and the winner will determined based on the following point system: 3 points for winning, 1 point for a draw, 0 points for a loss. If the bot crashes and are unable to recover, the game will continue and the bot will simply timeout and not do anything during their turn.

Each match will using the following configurations:

Group submissions are allowed but each person can only be part of one bot submission.


Bot Bowl III is funded by IEEE Computational Intelligence Society and and has a prize pool of $1,500.

IEEE CIS Sponsored Prizes: Sponsored Prize:

The machine learning prize sponsored by of $500 will be given to the most impressive bot that relies on machine learning or search algorithms. A jury of AI Researchers from will decide the winner of this prize based on the competition results, the implementation, and the submitted descriptions of the bots.

New Features

Pathfinding-assisted Move Actions

In previous Bot Bowls, bots could only move to adjacent squares, step by step. In Bot Bowl III, however, the FFAI framework will provide bots with available move actions to all reachable squares together with the safest and shortest path to reach it. Bots are then able to perform move actions to squares that are further away than one step, and the framework will perform all the intermediate steps automatically.

Available actions with pathfinding-assistance:


The user interface is, for demonstration purposes, showing here the safest path to blitz the opponent safety lineman. All green squares are reachable and each of them has a pre-computed path.

Available actions without pathfinding-assistance:


Previously, bots could only move players one square at the time

During a Blitz action, the set of available actions will also include pathfinding-assisted move actions to opponent players, where it will optimize the number of block rolls as well as the risk of moving. Similarly, pathfinding-assisted move actions are given for handoff and foul actions.

Check out our new tutorial on how to use pathfinding-assistance in reinforcement learning.

Forward Model

Previously, it was difficult to get a fast forward model up and running in FFAI due to the reliance on the slow copy.deepcopy() function. Thanks to amazing work by Mattias Bermell, FFAI now has a built-in forward model that is reasonably fast. At least much faster that what we had before!

It works by tracking changes to non-immutable properties in the game state. Such changes can then be reverted to go back in time, e.g. to reset the state, where we had to completely reinstantiate the entire game object before.

Check out our tutorials on how to use the forward model and how to implement a search-based bot.

Get Started

If you are completely new to bot development for Blood Bowl, there are plenty of detailed tutorials to get started. Please also join the Fantasy Football AI Discord server for news, discussions, and assistance if you get stuck.

Submit to Bot Bowl III

Submission details can be seen here.