Project AI year 2 – Assignment for Human Ambience Stream
The task of the current project is to support a human performing a demanding task involving visual attention, by designing an intelligent ambient system. In this project you will apply and integrate the techniques you have learned from a number of courses in the Artificial Intelligence curriculum to design an intelligent human ambience application. More specifically, you will learn how to specify domain knowledge in dynamic domain modeling, how to specify a design of an ambient agent for a particular domain using components for analysis, support, and parameter adaptation, and finally, how to apply all these in a specific software environment.
In this project, the challenge is to analyse a demanding task within a computer-simulated environment Missile Command and provide support for a human (player) performing this task.
Description of the domain
In the simulation environment used in this project the player faces objects falling from the top to the bottom of the screen. The player’s task is to classify each object as ally or enemy, and shoot down the enemies while letting allies land safely. See Appendix A for a detailed description of the game and the link where you can find it. The game imposes high demands on the player’s attention. The player may be supported by an ambient agent, which keeps track of how much attention the player has for the different objects, and provides some assistance in case the attention deviates too much from what it should be. Domain knowledge in the form of a dynamic model for attentional processes is described in Appendix B.
The project consists of 4 stages, namely conceptual design, and detailed design of integrated agent model (in LEADSTO), performing experiments within the provided software environment, and analysis of the results.
Assignment 1: Conceptual Design of the Agent Model (2 June – 4 June)
The course Integrative Modelling 1 presented a methodology for the design of ambient agents. One of the applications used as an example was the design of an ambient agent to support humans when performing demanding tasks thereby taking the attention of humans into account (Chapter 7). Different components were presented of an ambient agent model providing support for a human during these demanding tasks. In these components, a domain model is integrated. The first step in this assignment is to create a conceptual design of an ambient agent that assists a human performing the task within the Missile Command environment, and possibly intervenes in order to affect the human’s attention. The model should incorporate at least the following components:
· A dynamic domain model for attentional processes of the player. Knowledge for such a model can be found in Appendix B;
· An analysis model for attentional processes tuned to the particular case of the Missile Command environment;
· A support model for the player manipulating the player’s attention so as to enable better performance. You should invent helpful, effective support elements; an example which is not particularly helpful: the ambient agent sends an sms to the player’s mobile phone stating ‘have a look at coordinates (53,186)’;
· A model for adaptation of the parameters of the above models.
The format in which the model is described is to be adopted from the course Integrative Modelling 1, of which an example can for instance be seen in Chapter 7.
Assignment 2: Detailed Design of the Agent Model (5 June-10 June)
Here a detailed design of the agent model is specified in LEADSTO. The detailed specification should be delivered in full, and the details of it should be related to your model in Assignment 1, and some simulation traces have to show that it can work.
Assignment 3: Experimentation and adaptation (11 June- 21 June)
In this phase the results of Assignments 1 and 2 are the basis for an implementation within the specific software environment offered. This software environment connects to the simulation environment Missile Command and a TOBII eye tracker (see detailed description in Appendix C and D). For each group a computer equipped with this eye-tracker is available. The students will play the part of the human player. The experimentation phase consists of two parts:
One participant plays the game without support as it is given in Appendix A and his/her game process (including score and costs) is logged.
In order to enable an integration of the ambient agent model into the Missile Command environment, the Missile Command environment and a TOBII eye-tracking device will be provided together with a blueprint for the analysis, support and parameter adaptation model. Installation and calibration instructions for the software and TOBII hardware are found in Appendices C and D respectively. The provided blueprint for analysis, support and parameter adaptation has to be tuned to the detailed agent model resulting from Assignment 2, to reflect the specific characteristics, parameter values and other settings considered in Assignment 1. Play the game with support by the ambient agent model. Compare the scores you have obtained in Part 1 with newly obtained scores. Document your observations in the final report.
Continue your experiments changing characteristics and values of parameters in order to see whether you can achieve better support (i.e., higher scores) than with the first run. Document your observations in the final report. If you find out that further experimenting with the settings gives you better results than those of the first run, make adjustments in your documentation of Assignment 1 and 2.
In this phase you analyse usefulness of support/advice. Does a real player perform better when support is provided by the ambient agent? How much better? Conclusions are to be written down in a separate section of the final report.
Final Report
Your final report should include:
APPENDIX A. Description of the Missile Command game
APPENDIX C. Installation and Description of software (to be supplied after the 2nd week of the project)
APPENDIX D. Installation and Description of TOBII (to be supplied after the 2nd week of the project)