James Tulip

Towards adaptive online RTS AI with NEAT

Real Time Strategy (RTS) games are interesting from an Artificial Intelligence (AI) point of view because they involve a huge range of decision making from local tactical decisions to broad strategic considerations, all of which occur on a densely populated and fiercely contested map. However, most RTS AI used in commercial RTS games are predictable and can be exploited by expert players. Adaptive or evolutionary AI techniques offer the potential to create challenging AI opponents. Neural Evolution of Augmenting Technologies (NEAT) is a hybrid approach that applies Genetic Algorithm (GA) techniques to increase the efficiency of learning neural nets. This work presents an application of NEAT to RTS AI. It does so through a set of experiments in a realistic RTS environment. The results of the experiments show that NEAT can produce satisfactory RTS agents, and can also create agents capable of displaying complex in-game adaptive behavior. The results are significant because they show that NEAT can be used to evolve sophisticated RTS AI opponents without significant designer input or expertise, and without extensive databases of existing games.

Optimization Using Boundary Lookup Jump Point Search

Cache-based path-finding algorithms lose much of their advantage in dynamic environments where fast online search algorithms are required. Jump point search (JPS) is such a fast algorithm. It works by eliminating most map nodes from evaluation during path expansion. Boundary lookup jump point search (BL-JPS) is a modification that improves the speed of JPS. BL-JPS records the positions of obstacle boundaries and uses these via direct lookup to eliminate much of the iteration involved in searching for jump points in the JPS algorithm. Two sets of experiments are presented, demonstrating the effects of BL-JPS in both static and dynamic environments. The effects of different approaches to cache rebuilding for JPS+ in dynamic environments are also evaluated. Results show that BL-JPS is generally much faster than JPS. It is slower than JPS+ in static environments, but in dynamic environments, BL-JPS outperforms JPS+ for a single search. When multiple paths are searched, the effects of cache rebuilding gradually dominate the effects of search speed, resulting in JPS+ again becoming faster. However, combining JPS+ with BL-JPS provides a very fast path-finding algorithm (BL-JPS+) that outperforms JPS+ over a range of map types and numbers of paths searched.

Detecting flow in games using facial expressions

Many games use dynamic difficulty adjustment (DDA) to promote the achievement of flow and consequent positive affective states. However, performance based DDA assume a specific ludic attitude: that of the hard-core gamer. An alternative approach is to apply affective computing techniques to monitor players adjust difficulty to achieve the desired affective state directly. Such an emotion-controlled dynamic difficulty adjustment (EC-DDA) system might be more flexible and achieve better outcomes for a wider variety of players. Current approaches to monitoring affective state such as ECGs or EEGs can be very intrusive. However, monitoring affective state using facial expressions is non-intrusive, and can be done with minimal, generally existing hardware. This paper presents a simple arcade styled game incorporating a webcam and COTS facial expression analytical software. It presents the results of a set of experiments investigating the issues involved in collecting and analyzing facial expressions to determine player affect. Results demonstrate the feasibility of using facial expressions as a mechanism for determining player affect, but also illustrate some of the difficulties inherent in the EC- DDA approach. Specifically, the affects observed are not consistent with a standard interpretation of flow as characterized by high arousal and positive affect. Instead, even in a state of flow, the affect expressed may be flat. In other instances, affect may be highly variable, expressing a range of transitory basic emotions. Preliminary findings support the notion of flow as a complex cognitive state resulting from a cycle of transitions between simple affective states such as frustration and joy.