Decision-theoretic sequential decision making (SDM) is concerned with endowing an intelligent agent with the capability to choose the 'best' actions, i.e., those that that optimize the agent's performance on its task. SDM techniques have the potential to revolutionize many aspects of society, and recent successes, e.g., agents that learn to play Atari games and beat master Go players, have sparked renewed interest in this field.
However, despite these successes, fundamental problems of scalability prevent SDM methods from addressing other problems with hundreds or thousands of state variables. To overcome this barrier, INFLUENCE will develop a new class of `influence-based SDM methods' that address scalability issues by using novel ways of abstraction, thus making an important step towards realizing the promise of autonomous agent technology.
Staff involved: Frans A. Oliehoek.
The M2MGrids project aims to develop a horizontal platform in which physical sensors and devices can communicate with IT systems to allow for smart information exchange. A key business case within the project involves the research and development of automated dynamic power systems to allow for a smart flow of energy and data within the energy grid. As an example, imagine a scenario where a high power consumption is measured for a certain district of consumer homes. A present-day solution would be to raise the production of power to balance the consumption and production of power. However, a smart solution would be to re-schedule certain flexible devices within the home of a consumer (like a dishwasher or a washing machine) such that this problem dissolves by merely using devices and data in a smart way. Within the M2MGrids project, Delft is using their negotiation theory expertise to research and develop platforms which can be used within future smart homes to enable this smart scheduling of flexible devices. By creating platforms that enables a more effective and efficient use of energy, Delft is paving the way for a better and greener future!
Staff involved: Catholijn M. Jonker, Koen V. Hindriks.
The ReJAM project aims to develop Robots engaging elderly in Joint Activities with Music. The objective of ReJAM is the promotion of physical, cognitive, emotional and social wellbeing through various music-related activities, e.g. physical exercises, games, reminiscence, and making music together. The activities are designed especially for group activities: ReJAM is well-suited for use in meeting centers or together with visitors at home.
Staff involved: Mark A. Neerincx.
The overall aim of the project is to develop a reasoning framework that combines logic and quantitative techniques for Socially Adaptive Electronic Partners (SAEPs) that adapt their behavior to norms and values of people. This becomes more and more important as technology becomes an integral part of our daily lives. The computational reasoning techniques are aimed at determining when and to what extent norm-compliance can be guaranteed, and deciding what to do if in exceptional situations a norm cannot or should not be complied with.
Staff involved: M. Birna van Riemsdijk (PI), Catholijn M. Jonker.
The PAL (Personal Assistant for healthy Lifestyle) project proposal for Horizon 2020 was “favourably evaluated” and started on the 1st of March 2015 (EU grant is 4.5M Euro; ref. H2020-PHC-643783). This 4 year project involves the research partners TNO (coordinator), DFKI, FCSR, Imperial and Delft University of Technology, next to end-users (the hospitals Gelderse Vallei and Meander, and the Diabetics Associations of Netherlands and Italy), and SME’s (Mixel and Produxi). PAL will use, refine and extend the knowledge-base and support models of ALIZ-E to improve child’s diabetes regimen by assisting the child, health professional and parent. The PAL system will be composed of a social robot (NAO), its (mobile) avatar, and an extendable set of (mobile) health applications (diabetes diary, educational quizzes, sorting games, etc.), which all connect to a common knowledge-base and reasoning mechanism.
Staff involved: Mark A. Neerincx, Koen V. Hindriks, Joost Broekens.
Staff involved: Joost Broekens.
How does it feel to be a learning robot?
The aim in this project is that learning robots and virtual characters can express their own, and interpret human emotions in the context of their learning process, effectively making that process (1) more understandable and (2) better steerable by people. To achieve this, we develop a domain independent computational model of emotion based on reinforcement learning. Reinforcement learning is a successful computational model for skill learning. Intelligent robots and virtual characters can use reinforcement learning to learn the best strategy for executing a skill by trial and error and positive and negative feedback. However, there is no computational model of how emotions emerge from this learning process, and therefore no way to send out social signals to a human observer. We will extend the reinforcement learning model to a model of emotion by using neuroscientific and psychological insights into the relation between emotion and skill learning.
Staff involved: Joost Broekens.
Robots can be useful members of a rescue team in case of a disaster, but only if they do not burden the humans with complex controls. In my work I search for ways to make robots good team-members, so they automatically know where and to whom they can be of use. One of my methods is to systematically sabotage robot communication to find out what are good strategies to recover from this.
Staff involved: Koen V. Hindriks, Mark A. Neerincx.
This project studies how effectively and in what manner a stand-alone, multi-modal memory restructuring (3MR) system and Internet-based guided self-therapy version could be used for the treatment of post-traumatic stress disorder patients (PTSD).
Staff involved: Mark A. Neerincx, Willem-Paul Brinkman.
Negotiation is a complex emotional decision-making process aiming to reach an agreement to exchange goods or services. Although a daily activity, few people are effective negotiators. Existing support systems make a significant improvement if the negotiation space is well-understood, because computers can better cope with the computational complexity. However, the negotiation space can only be properly developed if the human parties jointly explore their interests. The inherent semantic problem and the emotional issues involved make that negotiation cannot be handled by artificial intelligence alone, and a human-machine collaborative system is required. We are developing a new type of human-machine collaborative system that combines the strengths of both and reduces the weaknesses. Fundamental in these systems will be that user and machine explicitly share a generic task model. Furthermore, such systems are to support humans in coping with emotions and moods in human-human interactions. For this purpose we will contribute new concepts, methods and techniques. For integrative bargaining we will develop such a system, called a Pocket Negotiator, to collaborate with human negotiators. The Pocket Negotiator will handle computational complexity issues, and provide bidding- and interaction advice, the user will handle background knowledge and interaction with the opponent negotiator.
Staff involved: Joost Broekens, Koen V. Hindriks, Catholijn M. Jonker.