i-PROGNOSIS is developing games to empower people living with Parkinson’s and sustain their quality of life. These games are intelligently configured by an Adaptation Manager, developed by our project partner CERTH. We tell you more about it in this article!
The Adaptation Manager is responsible for dynamically recommending game adaptations to users. It intends to deliver personalised and optimised advice in terms of game adaptations to keep players engaged. It supports both offline and online adaptation mechanisms; the reason for having two distinct mechanisms is to process different information about the players that may affect their performance.
- The offline adaptation mechanism takes into account player profiles and its ultimate goal is to match each player to the proper level of difficulty (LoD), i.e., to provide the most suitable challenge to the game.
- The online adaptation considers real-time player data related to player’s engagement and aims to select at specific time instances of the game the most appropriate dynamic element.
The offline adaptation mechanism is realized in the loading phase of the game and its purpose is to select the most appropriate game conditions for each player. These conditions are referred to as “levels of difficulty” and each one of them offers a different degree of challenge to the game.
The matching of a player with a scenario is based on a ranking system for the players and the scenarios. The selected scenario is used for the initialization of the game. In detail, for the modelling of player’s ability, we apply an Elo rating system and we adopt a similar approach for the rating of different LoD extending the work of Klinkenberg et al. For the selection of the appropriate scenario for each player, we apply pairwise comparisons between players and scenarios according to the equation.
Online adaptation is realized during the actual game play. Its goal is to select the proper dynamic game elements contributing to the enhancement of the player’s engagement. For the estimation of the player’s engagement, we are based on our recent work on player’s engagement recognition in serious games (Psaltis et al, 2017).
More specifically, we initially estimate the effective dimension of engagement by applying body motion and facial expression analysis and, subsequently, we extract features related to players’ cognitive and behavioural engagement based on the analysis of their interactions with the game. Finally, an artificial neural network is used for the automatic recognition of player’s engagement. Online adaptation considers real-time player data concerning player’s engagement estimation during specific time intervals in the game and selects each time the appropriate dynamic element (e.g. text messages, sounds or graphics) from a pool of elements realizing positive reinforcement or/and corrective feedback.
The collection of the engagement recordings for every element of the game constitutes the player’s engagement profile. To estimate the rating of each dynamic game element, that is the player’s preference model, we apply a discount factor approach that can be found in the context of Reinforcement Learning. Finally, for the selection of the appropriate element, we apply an ε-decreasing approach, which is a variation of the well-known epsilon-greedy algorithm.
In addition, an exer-motion game has been developed to integrate and test the proposed adaptation manager. The game consists of four different level of difficulties and two different types of dynamic elements (positive reinforcement and corrective feedback).
Figure 1. Four different Levels of Difficulty (LoDs) of the developed exergame
Figure 2. Two different types of dynamic elements (positive reinforcement and corrective feedback)
Avatar Authoring Tool
An avatar authoring tool has also been developed by CERTH, which allows the user to create personalized avatars by customizing different avatar features, e.g., clothes, skin color, hair style, color etc. The tool also supports a Facial Appearance Transfer technology that enables users to import their own facial appearance and structure onto the generated character.
Figure 3 Creation of a personalised avatar by the avatar authoring tool