Available thesis

1) Multimodal Emotion Recognition in HRI

Questa attività di tesi mira allo sviluppo di sistemi di machine learning per il riconoscimento delle emozioni da diversi segnali, quali ad esempio, l’espressione facciale, il battito cardiaco e la sudorazione.

2) Blueprint and dynamical persona for Personalization in HRI

A persona is defined as a single, specific hypothetical/fictitious person who represents a segment of the population with a realistic name, face, and description of their character (needs, goals, hopes, and attitudes). The Blueprint personas also include behavioural characteristics, which could affect both short and long-term success with interventions for managing a disease or adopting wellness. In this project, we want to explore how blueprint personas can be extended to include user characteristics that allow a robot to personalize its behaviour.

3) Showing Legible Behaviour

We will focus on identifying how a robot can show legible behaviour using cognitive and affective ToM information. We intend to model the non-verbal behaviours and interaction characteristics as meta-rules associated with the BDI logic used to infer the patient’s psychological and cognitive states of mind. This allows the robot to reason on their legibility considering the robotic hardware constraints (e.g., absence of a face to express facial emotions, or no-arms to gesture), and meeting people’s social constraints and preferences. Since the role of the robot (i.e., doctor, peer, carer, friend) and situational context (i.e., task type, risk, effects) affect people’s perception of the robot, factors that influence the legibility of a robot’s behaviour (i.e. people’s feeling of safety, comfort, efficiency and ability of the robot), have to be taken into account while deciding on the verbal and non-verbal cue to be used in the interaction.

4) Food recommendations via preference modelling

TBC

1) Knowledge-aware Inference for Speech Recognition in HRI

Previous works on knowledge-aware ASR inference have explored contextual and non-contextual (long-term) knowledge separately and have only experimented with English [1, 2]. The following can extend the previous works.
• Combining contextual and static knowledge for knowledge-aware inference of a pre-trained speech recognition model for HRI.
• A study to analyze effects of contextual speech recognition in HRI, possibly for a language other than English (e.g., Italian ASR).

[1]  P. Pramanick and C. Sarkar, “Can visual context improve automatic speech recognition for an embodied agent?” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, 2022, pp. 1946–1957.

[2]  ——, “Utilizing prior knowledge to improve automatic speech recognition in human-robot interactive scenarios,” in Companion of the 2023 ACM/IEEE International Conference on Human-Robot Interaction, 2023, pp. 471–475.

2) Emotional Response as Implicit Feedback for Confidence Estimation of Classifiers

The confidence estimate of a classifier is generally based on an estimation of how much the input matches the training data distribution. However, the emotional response of a co-located human who observes a prediction by a robot, can be used to regularize the model’s confidence estimation. For example, a positive valance may indicate a correct prediction, even though the model is not confident.

1) Integrazione di sistemi di ragionamento ad agenti in sistemi robotici

Questa attività di tesi mira all’integrazione di framework per la creazione di sistemi ad agenti tramite (ad esempio JASON and agentSpeak) in robot fisici (robot TEMI). Il robot dovrà essere in grado di pianificare ed eseguire una serie di attività di asseistenza per pazienti anziani in ambito domestico. L’interfaccia del robot sarà programmata in Android.

2) User Preferences in Providing Explanations

Study of how to formalize and learn the level of explainability that a user requires, and experiment modalities of interaction between human and autonomous systems in relation to explainability.

1) Computational Models for Nudging in HRI

This project investigates how a social robot can use nudging techniques for behavioural change. In this direction, verbal and non-verbal behaviours will be investigated to highlight the role of embodiment in nudging.