|Rank||Username||Country||Description||GECCO 2016||CIG 2016||Total|
Controller description: MaastCTS2
MCTS-based agent with a number of enhancements. Links with more information will be provided in the MaastCTS2 profile later, or can be requested by email (d (DOT) soemers (AT) gmail (DOT) com)
Controller description: adrienctx
In short, an open loop tree search. It builds a tree representing sequences of actions. At every iteration, one node/action is added to the tree. When navigating in the tree, the forward model is always used to generate states (i.e. no state is stored in the tree, which is why I called this "open loop"). To balance exploration/exploitation, it uses a classic UCB formula, with a small addition: a "taboo bias", that gives penalty to actions that lead to avatar positions visited in the recent past.
Controller description: Number27
Using a GA for local movement and a value map to direct the player across the level. The value map is created by evaluating each object type which results in a certain influence across the map. Notable events, the time spent in one area or non deterministic movements influence the players behaviour or how it chooses its optimal action. For deterministic games BFS is used.
Controller description: NovTea
We use a safety check to avoid close danger. We use a variant of the algorithm IW (Iterated Width) that combines IW1 with IW2.
Controller description: NovelTS
Optimized IW algorithm
Controller description: muzzle
A genetic algorithm for planning.
Controller description: CatLinux
A variant of GA for primary test.
Controller description: YOLOBOT
BFS in deterministic games. MCTS in other cases. A analysis of the current state is done. A list of all reachable itypes is created with the nearest sprite per itype. Given this list and the current state, empirical values determine the most interesting target. MCTS is used to get closer to the target without loosing the game.
Controller description: Return42
Depending on the type of game we use either a GA with custom heuristic or random walks. In deterministic puzzle games, we use a solver based on a customized A-Star algorithm.
Controller description: ICELab
In deterministic games ,we use Best First Search. In other cases ,we use a method that are based MCTS.
- Go Home -