EmoPlan

Automated Planning Models for Emotion-Aware Decision Making in Videogames

This project tries to get an insight on how to apply Automated Planning on videogames and storytelling. This project is funded by VisionDenmark, and it is done in collaboration with industrial partners: GenNarrate Aps.

Usually in AI Planning, our goal is to reach optimal solution for the problems we are dealing with, minimazing the costs of the actions. However, in narrative scenarios like story-driven videogames, the believability of autonomous Non-Player Characters (NPCs) relies more on their emotional and social behaviour than on always choosing the optimal action. One must be careful when applying planning in the design of "NPCs", so that their behaviour does not look unexplainable and emotionless.

The first step, is therefore, how to incorporate emotions and social relations into the planning task, so that they play a role on the behaviour of the NPCs. To do this, we will build on top of the computer model of the human decision-making process developed by GenNarrate ApS, which is based on research studies in Evolutionary Psychology and Cognitive Emotion Theory. In their model, emotions help determine the goals and actions of videogame NPCs.

The addition of all supplementary information regarding the "emotions and feelings" of the characters may have an overhead in performance if it not done carefully. We will analyse different ways of managing this information to achieve and efficient performance. At the same time, we try to establish how to evaluate the quality of the plans, and their relation to Cognitive Emotion Theory.

The results of this project was published at the 2022 Workshop on Scheduling and Planning Applications woRKshop (SPARK). The paper can be read here. You can also see the talk!

ReRoPro

Re-Use of Robotic-data in Production through search, simulation and learning

This is an Explorer-Project to investigate the underlying problems of data re-use on assembly in production. In contrast to other fields in AI, the potential of exploiting large data collections is not realized in robotics yet.

This project is a collaborative project among universities (SDU, KU and AAU), several companies (Norvo Nordisk and Nordbo) and cluster organization and research clusters (Odense Robotics and MADE).

You can find more information about this project in the following link

Projects at Saarland University

Minecraft-Saar

Within the SFB1102 about "Information Density and Linguistic Encoding" at Saarland university, we develop the project "Minecraft instruction giving project".

The project is a collaboration between the Foundations of Artificial Intelligence Group and the Department of Language Science and Technology at Saarland University. In a big picture, it consist of applying planning techniques in Natural Language Generation. I was part of the team while I was in Saarland University; but the project is still on going!

The Minecraft instruction giver

The objective of the project is to develop a system which generates effective technical instructions for the computer game “Minecraft”. The system makes a plan and communicates it to the listener, who is implementing the instructions. When the instructions are not followed or the user failed to do the action, the system replans the instructions to complete the task.

Example of the MC-Saar-Instruct

We aim to develop a system which is able to adapt the instructions to the previous knowledge of the listener. For example, in Minecraft, players with previuos experience already know where the blocks are; and therefore, they don't need any instruction to find them in the game interface, while someone who never played the game before would need further information about it. Following the same reasoning, the listeners learn while they performs the actions, so they will usually need less instructions at the end of the task than at the beginning.

The Minecraft environment has been chosen because it lets us to easily perform experiments, measure the efficiency/difficulty of the actions and instructions and the effectiviness of the system. But the future applications of this research go much further. We pursue to create a rational speaker who trades off communicative success (being understood by the listener) against succinctness, in particular with respect to the listener’s domain knowledge and the division of labor between language and video.

You can find more information on the project, the related publications and the current prototype at the webpage of Minecraft Saar.