[DEV view]:
Each gladiator has 4 neural networks that control 1) facing; 2) movement; 3) attacking; 4) map boundary
Each AI, in it's own way, will learn to understand where the map boundary is, how to move toward it's opponent, how to face it's opponent, and when to swing/not swing.
Attributes for each gladiator:
- speed (initiative)
- durabilty (abilty to take damage)
- strength (ability to deal damage)
Very simple to start with. Players will, initially, buy rookies based on the above. As the gladiators battle and the neural networks learn, players need to judge if they've bought a lemon or not. A gladiator with good attributes but a slow neural network may work out in the long game.
All rookies know nothing - not even know how to move - so initial battles are ... entertaining ... as the AI begins to understand it's purpose. A gladiator winning it's first round moves to a new pool of other gladiators that have also won a single round. This means victors play victors and not rookies. A gladiator winning a 2nd round moves to a new pool of other gladiators that have won 2 rounds and so on. eg. gladiators that have one 5 rounds will battle gladiators that have also won 5 rounds. Equality and balance is not assured - but competitiveness is.
Losing around doesn't mean death - but that is always a possibility. AI that learns to attack from behind have a greater chance of landing a fatal and terminal blow. Losing, generally, means the gladiator remains in it's current pool but death is death and a part of a gladiators life.
Build up your stable of gladiators. Watch them learn and grow. Watch the hero's evolve. Enjoy learning about neural networks and genuine machine learning. Run auto-battles in the background and become czar of gladiator battles.