Search

Pattern recognition in EMG-based control systems suffer from increase in error rate over time, which could lead to unwanted behavior. This so-called concept drift in myoelectric control systems could be caused by fatigue, sensor replacement and varying skin conditions. To circumvent concept drift, adaptation strategies could be used to retrain a pattern recognition system, which could lead to comparable error rates over multiple days. In this study, we investigated the error rate development over one week and compared three adaptation strategies to reduce the error rate increase. The three adaptation strategies were based on entropy, on backward prediction and a combination of backward prediction and entropy. Ten able-bodied subjects were measured on four measurement days while performing gait-related activities. During the measurement electromyography and kinematics were recorded. The three adaptation strategies were implemented and compared against the baseline error rate and against adaptation using the ground truth labels. It can be concluded that without adaptation the baseline error rate increases significantly from day 1 to 2, but plateaus on day 2, 3 and 7. Of the three tested adaptation strategies, entropy based adaptation showed the smallest increase in error rate over time. It can be concluded that entropy based adaptation is simple to implement and can be considered a feasible adaptation strategy for lower limb pattern recognition.

At vital moments in professional soccer matches, penalties were often missed. Psychological factors, such as anxiety and pressure, are among the critical causes of the mistakes, commonly known aschoking under pressure. Nevertheless, the factors have not been fully explored. In this study, we used functional near-infrared spectroscopy (fNIRS) to investigate the influence of the brain on this process. Anin-situstudy was set-up (N= 22), in which each participant took 15 penalties under three different pressure conditions: without a goalkeeper, with an amiable goalkeeper, and with a competitive goalkeeper. Both experienced and inexperienced soccer players were recruited, and the brain activation was compared across groups. Besides, fNIRS activation was compared between sessions that participants felt anxious against sessions without anxiety report, and between penalty-scoring and -missing sessions. The results show that the task-relevant brain region, the motor cortex, was more activated when players were not experiencing performance anxiety. The activation of task-irrelevant areas was shown to be related to players experiencing anxiety and missing penalties, especially the prefrontal cortex (PFC). More particularly, an overall higher activation of the PFC and an increase of PFC lateral asymmetry were related to anxious players and missed penalties, which can be caused by players' worries about the consequences of scoring or missing the penalty kicks. When experienced players were feeling anxious, their left temporal cortex activation increased, which could be an indication that experienced overthink the situation and neglect their automated skills. Besides, the left temporal cortex activation is higher when inexperienced players succeeded to score a penalty. Overall, the results of this study are in line with the neural efficiency theory and demonstrate the feasibility and ecological validity to detect neurological clues relevant to anxiety and performance from fNIRS recordingsin the field.

Autonomous robots require high degrees of cognitive and motoric intelligence to come into our everyday life. In non-structured environments and in the presence of uncertainties, such degrees of intelligence are not easy to obtain. Reinforcement learning algorithms have proven to be capable of solving complicated robotics tasks in an end-to-end fashion without any need for hand-crafted features or policies. Especially in the context of robotics, in which the cost of real-world data is usually extremely high, reinforcement learning solutions achieving high sample efficiency are needed. In this paper, we propose a framework combining the learning of a low-dimensional state representation, from high-dimensional observations coming from the robot's raw sensory readings, with the learning of the optimal policy, given the learned state representation. We evaluate our framework in the context of mobile robot navigation in the case of continuous state and action spaces. Moreover, we study the problem of transferring what learned in the simulated virtual environment to the real robot without further retraining using real-world data in the presence of visual and depth distractors, such as lighting changes and moving obstacles., Comment: Paper Accepted at IROS2021. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

Human Activity Recognition is focused on the use of sensing technology to classify human activities and to infer human behavior. While traditional machine learning approaches use hand-crafted features to train their models, recent advancements in neural networks allow for automatic feature extraction. Auto-encoders are a type of neural network that can learn complex representations of the data and are commonly used for anomaly detection. In this work we propose a novel multi-class algorithm which consists of an ensemble of auto-encoders where each auto-encoder is associated with a unique class. We compared the proposed approach with other state-of-the-art approaches in the context of human activity recognition. Experimental results show that ensembles of auto-encoders can be efficient, robust and competitive. Moreover, this modular classifier structure allows for more flexible models. For example, the extension of the number of classes, by the inclusion of new auto-encoders, without the necessity to retrain the whole model.

In this research, we investigate the use of Reinforcement Learning (RL) for an effective and robust solution for exploring unknown and indoor environments and reconstructing their maps. We benefit from a Simultaneous Localization and Mapping (SLAM) algorithm for real-time robot localization and mapping. Three different reward functions are compared and tested in different environments with growing complexity. The performances of the three different RL-based path planners are assessed not only on the training environments, but also on an a priori unseen environment to test the generalization properties of the policies. The results indicate that RL-based planners trained to maximize the coverage of the map are able to consistently explore and construct the maps of different indoor environments.

Full-body motion capture typically requires sensors/markers to be placed on each rigid body segment, which results in long setup times and is obtrusive. The number of sensors/markers can be reduced using deep learning or offline methods. However, this requires large training datasets and/or sufficient computational resources. Therefore, we investigate the following research question: &ldquo, What is the performance of a shallow approach, compared to a deep learning one, for estimating time coherent full-body poses using only five inertial sensors?&rdquo, We propose to incorporate past/future inertial sensor information into a stacked input vector, which is fed to a shallow neural network for estimating full-body poses. Shallow and deep learning approaches are compared using the same input vector configurations. Additionally, the inclusion of acceleration input is evaluated. The results show that a shallow learning approach can estimate full-body poses with a similar accuracy (~6 cm) to that of a deep learning approach (~7 cm). However, the jerk errors are smaller using the deep learning approach, which can be the effect of explicit recurrent modelling. Furthermore, it is shown that the delay using a shallow learning approach (72 ms) is smaller than that of a deep learning approach (117 ms).

Contains fulltext : 135011.pdf (Publisher’s version ) (Closed access) Visual attention is used to selectively filter relevant information depending on current task demands and goals. Visual attention is called object-based attention when it is directed to coherent forms or objects in the visual field. This study used real-time functional magnetic resonance imaging for moment-to-moment decoding of attention to spatially overlapped objects belonging to two different object categories. First, a whole-brain classifier was trained on pictures of faces and places. Subjects then saw transparently overlapped pictures of a face and a place, and attended to only one of them while ignoring the other. The category of the attended object, face or place, was decoded on a scan-by-scan basis using the previously trained decoder. The decoder performed at 77.6% accuracy indicating that despite competing bottom-up sensory input, object-based visual attention biased neural patterns towards that of the attended object. Furthermore, a comparison between different classification approaches indicated that the representation of faces and places is distributed rather than focal. This implies that real-time decoding of object-based attention requires a multivariate decoding approach that can detect these distributed patterns of cortical activity. 11 p.

We propose a novel approach to developing a tractable affective dialogue model for probabilistic frame-based dialogue systems. The affective dialogue model, based on Partially Observable Markov Decision Process (POMDP) and Dynamic Decision Network (DDN) techniques, is composed of two main parts: the slot-level dialogue manager and the global dialogue manager. It has two new features: (1) being able to deal with a large number of slots and (2) being able to take into account some aspects of the user's affective state in deriving the adaptive dialogue strategies. Our implemented prototype dialogue manager can handle hundreds of slots, where each individual slot might have hundreds of values. Our approach is illustrated through a route navigation example in the crisis management domain. We conducted various experiments to evaluate our approach and to compare it with approximate POMDP techniques and handcrafted policies. The experimental results showed that the DDN–POMDP policy outperforms three handcrafted policies when the user's action error is induced by stress as well as when the observation error increases. Further, performance of the one-step look-ahead DDN–POMDP policy after optimizing its internal reward is close to state-of-the-art approximate POMDP counterparts.

The project Augmented Multi-party Interaction (AMI) is concerned with the development of meeting browsers and remote meeting assistants for instrumented meeting rooms – and the required component technologies R&D themes: group dynamics, audio, visual, and multimodal processing, content abstraction, and human-computer interaction. The audio-visual processing workpackage within AMI addresses the automatic recognition from audio, video, and combined audio-video streams, that have been recorded during meetings. In this article we describe the progress that has been made in the first two years of the project. We show how the large problem of audio-visual processing in meetings can be split into seven questions, like “Who is acting during the meeting?��?. We then show which algorithms and methods have been developed and evaluated for the automatic answering of these questions.

Advances in the field of touch recognition could open up applications for touch-based interaction in areas such as Human-Robot Interaction (HRI). We extended this challenge to the research community working on multimodal interaction with the goal of sparking interest in the touch modality and to promote exploration of the use of data processing techniques from other more mature modalities for touch recognition. Two data sets were made available containing labeled pressure sensor data of social touch gestures that were performed by touching a touch-sensitive surface with the hand. Each set was collected from similar sensor grids, but under conditions reflecting different application orientations: CoST: Corpus of Social Touch and HAART: The Human-Animal Affective Robot Touch gesture set. In this paper we describe the challenge protocol and summa- rize the results from the touch challenge hosted in conjunction with the 2015 ACM International Conference on Multi- modal Interaction (ICMI). The most important outcomes of the challenges were: (1) transferring techniques from other modalities, such as image processing, speech, and human action recognition provided valuable feature sets; (2) gesture classification confusions were similar despite the various data processing methods used. Categories

Touch is an important interaction modality in social interaction, for instance touch can communicate emotions and can intensify emotions communicated by other modalities. In this paper we explore the use of Neural Networks for the classification of touch. The exploration and assessment of Neural Networks (NNs) is based on the Corpus of Social Touch established by Jung et al. This corpus was split in a train set (65%) and test set (35%), the train set was used to find the optimal parameters for the NN and for training the final model. Also different feature sets were investigated; the basic feature set included in the corpus, energy-histogram and dynamical features. Using all features led to the best performance of 64% on the test set, using a NN consisting of one hidden layer with 46 neurones. The confusion matrix showed the expected high confusion between pat-tap and grab-squeeze. A leave-one-subject-out approach lead to a performance of 54%, which is comparable with the results of Jung et al.

Automatic sleep staging on an online basis has recently emerged as a research topic motivated by fundamental sleep research. The aim of this paper is to find optimal signal processing methods and machine learning algorithms to achieve online sleep staging on the basis of a single EEG signal. The classification performance obtained using six different EEG signals and various signal processing feature sets is compared using the kappa statistic which has very recently become popular in sleep staging research. A variable duration of the EEG segment (or epoch) to decide on the sleep stage is also analyzed. Spectral-domain, time-domain, linear, and nonlinear features are compared in terms of performance and two types of machine learning approaches (random forests and support vector machines) are assessed. We have determined that frontal EEG signals, with spectral linear features, epoch durations between 18 and 30 seconds, and a random forest classifier lead to optimal classification performance while ensuring real-time online operation.

The lazy caching protocol proposed by Afek, Brown and Merritt [ABM93], is explained and formally proven correct by means of compositional methods. The protocol is decomposed into four simple protocols, which are of interest on their own. A top level proof is given that is to a large extent independent of the particular model used for the more detailed proofs and allows for a number of generalizations of the original lazy caching protocol. Detailed proofs of safety and liveness properties are given using CSP and trace logic.

Fraud, waste, and abuse in the U.S. healthcare system are estimated at $700 billion annually. Predictive analytics offers government and private payers the opportunity to identify and prevent or recover such billings. This paper proposes a data-driven method for fraud detection based on comparative research, fraud cases, and literature review. Unsupervised data mining techniques such as outlier detection are suggested as effective predictors for fraud. Based on a multi-dimensional data model developed for Medicaid claim data, specific metrics for dental providers were developed and evaluated in analytical experiments using outlier detection applied to claim, provider, and patient data in a state Medicaid program. The proposed methodology enabled successful identification of fraudulent activity, with 12 of the top 17 suspicious providers (71%) referred to officials for investigation with clearly anomalous and inappropriate activity. Future research is underway to extend the method to other specialties and enable its use by fraud analysts.

Brain–computer interfaces (BCI) provide a valuable new input modality within human–computer interaction systems. However, like other body-based inputs such as gesture or gaze based systems, the system recognition of input commands is still far from perfect. This raises important questions, such as what level of control should such an interface be able to provide. What is the relationship between actual and perceived control? And in the case of applications for entertainment in which fun is an important part of user experience, should we even aim for the highest level of control, or is the optimum elsewhere? In this paper, we evaluate whether we can modulate the amount of control and if a game can be fun with less than perfect control. In the experiment users (n = 158) played a simple game in which a hamster has to be guided to the exit of a maze. The amount of control the user has over the hamster is varied. The variation of control through confusion matrices makes it possible to simulate the experience of using a BCI, while using the traditional keyboard for input. After each session the user completed a short questionnaire on user experience and perceived control. Analysis of the data showed that the perceived control of the user could largely be explained by the amount of control in the respective session. As expected, user frustration decreases with increasing control. Moreover, the results indicate that the relation between fun and control is not linear. Although at lower levels of control fun does increase with improved control, the level of fun drops just before perfect control is reached (with an optimum around 96%). This poses new insights for developers of games who want to incorporate some form of BCI or other modality with unreliable input in their game: for creating a fun game, unreliable input can be used to create a challenge for the user.

While a BCI usually aims to provide an alternative communication channel for disabled users who have difficulties to move or to speak, we focused on BCIs as a way to retrieve and use information about an individual’s cognitive or affective state without requiring any effort or intention of the user to convey this information. Providing only an extra channel of information rather than a replacement of certain functions, such BCIs could be useful for healthy users as well. We describe the results of our studies on neurophysiological correlates of attention, workload and emotion, as well as our efforts to include physiological variables. We found different features in EEG to be indicative of attention and workload, while emotional state may be better measured by physiological variables like heart rate and skin conductance. Potential applications are described. We argue that major challenges lie in hardware and generalization issues.

Brain-computer interfaces offer a valuable input modality, which unfortunately comes also with a high degree of uncertainty. There are simple methods to improve detection accuracy after the incoming brain activity has already been classified, which can be divided into (1) gathering additional evidence from other sources of information, and (2) transforming the unstable classification results to be more easy to control. The methods described are easy to implement, but it is essential to apply them in the right way. This paper provides an overview of the different techniques, showing where to apply them and comparing the effects. Detection accuracy is important, but there are trade-offs to consider. Future research should investigate the effectiveness of these methods in their context of use, as well as the optimal settings to obtain the right balance between functionality and meeting the user's expectations for maximum acceptance.

Recently, wireless sensor networks (WSNs) have become mature enough to go beyond being simple fine-grained continuous monitoring platforms and have become one of the enabling technologies for early-warning disaster systems. Event detection functionality of WSNs can be of great help and importance for (near) real-time detection of, for example, meteorological natural hazards and wild and residential fires. From the data-mining perspective, many real world events exhibit specific patterns, which can be detected by applying machine learning (ML) techniques. In this paper, we introduce ML techniques for distributed event detection in WSNs and evaluate their performance and applicability for early detection of disasters, specifically residential fires. To this end, we present a distributed event detection approach incorporating a novel reputation-based voting and the decision tree and evaluate its performance in terms of detection accuracy and time complexity.

Brain-computer interfaces (BCIs) are known to suffer from spontaneous changes in the brain activity. If changes in the mental state of the user are reflected in the brain signals used for control, the behavior of a BCI is directly influenced by these states. We investigate the influence of a state of loss of control in a variant of Pacman on the performance of BCIs based on motor control. To study the effect a temporal loss of control has on the BCI performance, BCI classifiers were trained on electroencephalography (EEG) recorded during the normal control condition, and the classification performance on segments of EEG from the normal and loss of control condition was compared. Classifiers based on event-related desynchronization unexpectedly performed significantly better during the loss of control condition; for the event-related potential classifiers there was no significant difference in performance.

Offline analysis pipelines have been developed and evaluated for the detection of covert attention from electroen-cephalography recordings, and the detection of overt attention in terms of eye movement based on electrooculographic measurements. Some additional analysis were done in order to prepare the pipelines for use in a real-time system. This real-time system and a game application in which these pipelines are to be used were implemented. The game is set in a virtual environment where player is a wildlife photographer on an uninhabited island. Overt attention is used to adjust the angle of the first person camera, when the player is tracking animals. When making a photograph, the animal will flee when it notices it is looked at directly, so covert attention is required to get a good shot. Future work will entail user tests with this system to evaluate usability, user experience, and characteristics of the signals related to overt and covert attention when used in such an immersive environment.

Recently, wireless sensor networks (WSNs) have become mature enough to go beyond being simple fine-grained continuous monitoring platforms and become one of the enabling technologies for disaster early-warning systems. Event detection functionality of WSNs can be of great help and importance for (near) real-time detection of, for example, meteorological natural hazards and wild and residential fires. From the data-mining perspective, many real world events exhibit specific patterns, which can be detected by applying machine learning (ML) techniques. In this paper, we introduce ML techniques for distributed event detection in WSNs and evaluate their performance and applicability for early detection of disasters, specifically residential fires. To this end, we present a distributed event detection approach incorporating a novel reputation-based voting and the decision tree and evaluate its performance in terms of detection accuracy and time complexity.

In the near future, brain-computer interface (BCI) applications for non-disabled users will require multimodal interaction and tolerance to dynamic environment. However, this conflicts with the highly sensitive recording techniques used for BCIs, such as electroencephalography (EEG). Advanced machine learning and signal processing techniques are required to decorrelate desired brain signals from the rest. This paper proposes a signal processing pipeline and two classification methods suitable for multiclass EEG analysis. The methods were tested in an experiment on separating left/right hand imagery in presence/absence of speech. The analyses showed that the presence of speech during motor imagery did not affect the classification accuracy significantly and regardless of the presence of speech, the proposed methods were able to separate left and right hand imagery with an accuracy of 60%. The best overall accuracy achieved for the 5-class separation of all the tasks was 47% and both proposed methods performed equally well. In addition, the analysis of event-related spectral power changes revealed characteristics related to motor imagery and speech.

Brain-Computer Interface (BCI) research has made great progress recently [1-3]. However, this progress has some negative side effects: growing fragmentation among different researchers, confusion about the best research directions, and ongoing disagreement over terms and definitions. Future BNCI is a Coordination and Support Action funded by the European Commission that aims to counteract these trends by helping new and existing researchers identify each other, encouraging effective collaborations, developing roadmaps and frameworks, and establishing standardized terminology. The knowledge developed in Future BNCI will be disseminated through conferences, workshops, journal publications, a book, and a website.

A model related to the CTS model of [KMP96] is introduced that deals with distributed systems with clocks that are not perfectly synchonized. The basis of the system is temporal logic of actions [Lam91], which ensures that the model is easily integrated with other formalisms. Verification rules and refinement rules for timed modules are proposed, and applied to a clock synchronization algorithm derived from [CAS86].

In this paper, we describe our investigation of traces of naturally occurring emotions in electrical brain signals, that can be used to build interfaces that respond to our emotional state. This study confirms a number of known affective correlates in a realistic, uncontrolled environment for the emotions of valence (or pleasure), arousal and dominance: (1) a significant decrease in frontal power in the theta range is found for increasingly positive valence, (2) a significant frontal increase in power in the alpha range is associated with increasing emotional arousal, (3) a significant right posterior power increase in the delta range correlates with increasing arousal and (4) asymmetry in power in the lower alpha bands correlates with self-reported valence. Furthermore, asymmetry in the higher alpha bands correlates with self-reported dominance. These last two effects provide a simple measure for subjective feelings of pleasure and feelings of control.

This paper covers part of a study which concerns the feasibility of real-time word spotting with partially recurrent neural networks (PRNN’s). PRNN’s have already proven appropriate for other examples of pure sequence recognition [1, 2]. However choices concerning architectural and learning aspects are still hard to make. One of the questions still to be answered, is how these aspects influence the term of memory of a PRNN. This paper tries to obtain some directives regarding architectures and learning algorithms.

In this paper a rule extraction method for a multilayered neural network is proposed. The method is not only able to extract rules from a neural network trained with boolean valued inputs but also from a neural network trained with continuous valued inputs.

An assertional proof system is proposed for a shared variable language extended with real-time constructs and synchronization, and a layering operator. The proof system is used to check real-time side conditions for an extended Communication Closed Layers rule. Reversely, this rule extends the scope of applicability of the proof system. The resulting system combines algebraic and assertional reasoning. A number of small examples are discussed.

We developed Wild Photoshoot, a game that uses naturally-occurring neurophysiological activity to augment the interaction in a virtual environment in an intuitive way. In this game, the user is a wildlife photographer. Besides normal movement controls (mouse and keyboard), the camera is adjusted according to where the user is looking (overt attention, OA). When the animal has been found, the user will have to use covert attention (CA) (Van Gerven et al., 2009) to take the picture, because when the user looks at the animal directly, it will flee. The mental tasks for OA and CA come naturally given the situation. Initial offline tests assessed the performance of EEG-based CA and EOG-based OA. For CA, the average accuracy was 67% (2 classes, 4 participants), with the pipeline: common average reference, band pass 8-14 Hz, whitening, covariance and logistic regression. The pipeline for OA is based on Barea et al., 2003 and Itakura and Sakamoto, 2010: band pass 0.05-20 Hz, derivation, threshold, integration, and linear regression. For horizontal eye movement the average error was 2.2cm, and for vertical eye movement 4.8cm (4 participants). Although BCIs are the last option for interaction for those patients who have no residual muscle control, there are also patients with limited control, who could benefit from a hybrid BCI setup which combines these two inputs. The naturalness of these inputs can make BCIs easy to use; an aspect that will be appreciated by both patients and healthy users.

A new notion of refinement and several other new operators are proposed that allow for a compositional algebraic characterization of action systems and serializability in distributed database systems. A simple design language is introduced and is provided with a semantics essentially based on partial order models.

Clinical decision support systems (CDSS) can support clinicians in selecting appropriate treatments for patients. The objective of this study was to examine if triaging patients with LBP to the most optimal treatment can be improved by using a data-driven approach with the help of machine learning technologies as base of such a CDSS. A clinical database of the Groningen Spine Center containing patient-reported data from 1546 patients with LBP was used. From this dataset, a training dataset with 354 features was labeled on eight different treatments actually received by these patients. With this dataset, models were trained. A test dataset with 50 cases judged on treatments by 4 experts in LBP triage to treatments was used to test these models with data not used to train the models. Prediction accuracy and average area under curve (AUC) were used as performance measures for the models. The AUC values indicated small to medium learning effects showing that machine learning on patient-reported data, to model decision-making processes on treatments for LBP, may be possible. One of the best performing models was the Bayesian Network (BN) model; e.g. predicted surgery with accuracy 0.78 (95% C.I. 0.68– 0.87) and AUC 0.70. Benefits to using BNs compared to other supervised machine learning techniques are that it is easy to exploit expert knowledge in BN models, meaning that advices generated by the model can be explained. The next step is to improve the BN so that it can be used in a CDSS.

Brain-computer interfaces (BCI) provide a valuable new input modality within human- computer interaction systems, but like other body-based inputs, the system recognition of input commands is far from perfect. This raises important questions, such as: What level of control should such an interface be able to provide? What is the relationship between actual and perceived control? And in the case of applications for entertainment in which fun is an important part of user experience, should we even aim for perfect control, or is the optimum elsewhere? In this experiment the user plays a simple game in which a hamster has to be guided to the exit of a maze, in which the amount of control the user has over the hamster is varied. The variation of control through confusion matrices makes it possible to simulate the experience of using a BCI, while using the traditional keyboard for input. After each session the user �lled out a short questionnaire on fun and perceived control. Analysis of the data showed that the perceived control of the user could largely be explained by the amount of control in the respective session. As expected, user frustration decreases with increasing control. Moreover, the results indicate that the relation between fun and control is not linear. Although in the beginning fun does increase with improved control, the level of fun drops again just before perfect control is reached. This poses new insights for developers of games wanting to incorporate some form of BCI in their game: for creating a fun game, unreliable input can be used to create a challenge for the user.

Data scientists are in high demand in many disciplines and domains. This paper describes the data science course open to all master students of the University of Twente. We outline the main challenges of teaching a large and heterogeneous population of non-computer science students about data science and how we addressed them, as well as a historical perspective on how the course grew and evolved.

We present a map-less path planning algorithm based on Deep Reinforcement Learning (DRL) for mobile robots navigating in unknown environment that only relies on 40-dimensional raw laser data and odometry information. The planner is trained using a reward function shaped based on the online knowledge of the map of the training environment, obtained using grid-based Rao-Blackwellized particle filter, in an attempt to enhance the obstacle awareness of the agent. The agent is trained in a complex simulated environment and evaluated in two unseen ones. We show that the policy trained using the introduced reward function not only outperforms standard reward functions in terms of convergence speed, by a reduction of 36.9\% of the iteration steps, and reduction of the collision samples, but it also drastically improves the behaviour of the agent in unseen environments, respectively by 23\% in a simpler workspace and by 45\% in a more clustered one. Furthermore, the policy trained in the simulation environment can be directly and successfully transferred to the real robot.

We propose a new approach to developing a tractable affective dialogue model for general probabilistic frame-based dialogue systems. The dialogue model, based on the Partially Observable Markov Decision Process (POMDP) and the Dynamic Decision Network (DDN) techniques, is composed of two main parts, the slot level dialogue manager and the global dialogue manager. It has two new features: (1) being able to deal with a large number of slots and (2) being able to take into account some aspects of the user's affective state in deriving the adaptive dialogue strategies. Our implemented dialogue manager prototype can handle hundreds of slots; each slot might have many values. A first evaluation of the slot level dialogue manager (1-slot case) showed that with a 95% confidence level the DDN-POMDP dialogue strategy outperforms three handcrafted dialogue strategies when the user's action error is induced by stress.