Robots' Inner Speech meets Thories of emotion: towards Affective Robotics in Human Robot Interaction

Engineering notes

Key topics: autonomous driving, deployment, planning, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Robotics is undergoing a paradigm shift, evolving from systems primarily designed to execute mechanical or repetitive tasks into autonomous agents capable of engaging in meaningful, socially coherent interactions with humans.As robots become increasingly integrated into everyday environments, ranging from domestic assistance to healthcare, education, and therapeutic contexts, the demand for machines capable of understanding, expressing, and regulating emotions has grown considerably.Emotional intelligence is no longer perceived as an optional layer for enhancing user experience; it is increasingly recognised as a core requirement for establishing trust, fostering long-term interaction, and enabling cooperation between humans and artificial agents.Nevertheless, implementing this capability computationally is a significant challenge that demands the integration of affective modeling, cognitive architectures, and communication systems.Emotions in biological organisms serve as a crucial mediator between cognitive processes and environmental demands.They influence decision-making, guide attention, facilitate memory consolidation, and play a key role in social bonding.From an evolutionary perspective, emotions act as adaptive responses that help organisms prioritise goals, evaluate risks, and coordinate behaviour within complex social structures.Translating these mechanisms into artificial systems involves not only developing mathematical models of affective states but also integrating them into architectures that enable flexible, context-sensitive behaviour.Many existing approaches in affective computing remain primarily reactive, mapping external stimuli to predefined emotional outputs, often yielding limited or shallow expressions that lack introspective depth.This doctoral research aims to address these limitations by introducing a novel paradigm that integrates computational emotional models with inner speech mechanisms.Inner speech, also known as self-talk or covert verbal thinking, is a welldocumented phenomenon in human cognition that contributes to self-regulation, planning, problem-solving, and emotional adjustment.In humans, inner speech enables the simulation of social dialogue within the self, allowing individuals to reason about Abstract their feelings, anticipate consequences, and regulate their reactions in real time.The central hypothesis of this dissertation is that endowing robots with an analogous internal dialogue mechanism can support a form of machine introspection that enhances both emotional coherence and transparency in interaction.This introspection lays the foundation for robot consciousness, explored in detail in Chapter 2.The research presented in this dissertation demonstrates that the integration of inner speech into computational emotional models provides a viable pathway toward more reflective and communicative robotic systems.Rather than focusing solely on emotional expression, the proposed approach enables robots to reflect upon, regulate, and communicate their internal states in a manner that is intelligible and socially appropriate for human partners.The findings suggest that such models can enhance the quality of human-robot interaction across multiple domains, while contributing to user trust and system acceptability.Future research will explore the scalability of this approach to more complex emotional repertoires, its integration with adaptive learning mechanisms, and its ethical implications, particularly regarding transparency, accountability, and the anthropomorphisation of artificial agents.The dissertation is organised to reflect both its theoretical foundations and its applied contributions.The initial chapters review the state of the art in affective computing, emotion modelling, and inner speech, and outline the methodological choices underlying the proposed computational models.The central chapters detail the system architecture, its integration into robotic platforms, and the results of controlled experimental evaluations.The final chapters focus on real-world applications, describing interaction design, evaluation criteria, and insights gained from deployment in practical settings.The first part of the research focuses on the design and formalisation of computational models capable of generating and modulating emotional states through mechanisms of inner speech.These models adopt a hybrid architecture that combines symbolic reasoning with dynamic affective processes, enabling a bidirectional flow between cognitive appraisals and emotional expressions.In Chapter 1, the thesis introduces the conceptual and computational integration of inner speech within an appraisal-based emotional model for robots.Building on appraisal theories, which describe emotions as emerging from cognitive evaluations of situational variables, this chapter examines how self-directed dialogue can serve as an internal mechanism that supports emotional computation.Inner speech is modelled as a structured process of self-reflection that enables the robot to focus on contextually relevant information, compute assessment variables, and generate emotionally coherent responses.Particular attention is devoted to analysing emotional dynamics under stressful conditions, demonstrating that the proposed model produces patterns Abstract consistent with those observed in healthy adults.Furthermore, the chapter discusses how the externalisation of inner reasoning through think-aloud behaviour can improve transparency and coordination in collaborative tasks, supporting clearer joint decisionmaking and mutual understanding between human and robot partners.In Chapter 2, the research extends the investigation from appraisal-based evaluation to a broader affective framework grounded in Damasio's theory, in which emotions arise from the dynamic interaction between bodily-like signals and cognitive processes.This chapter explores how a robot can move beyond mere detection or simulation of emotions toward a computational architecture that supports emotionally grounded responses mediated by inner speech.Self-directed dialogue is implemented as a mechanism that enables the robot to articulate its interpretation of contextual events and its internal state, thereby fostering the emergence of more coherent and interpretable behaviours.The model is deployed on a real robotic platform, and experimental findings demonstrate that human participants interacting with the system can perceive the robot's emotional states.The results highlight how integrating embodied-cognitive mechanisms with inner speech enhances perceived empathy, trust, and engagement in Human-Robot Interaction.Building on this foundation, the second part of the dissertation shifts focus to a preliminary investigation of the application of the developed computational models in a medical context.In Chapter 3, the thesis places the developed models in a complex, high-risk collaborative environment, with a specific focus on a medical application.The chapter examines the role of robotic inner speech in cooperation with a nurse during the preparation of a surgical table, a task that requires precision and careful instrument placement to avoid adverse procedural outcomes.The study analyses how the robot's self-directed dialogue contributes to reassurance and stress management in demanding contexts, while simultaneously enhancing clarity and understanding of task-related instructions.The findings demonstrate that inner speech retains its effectiveness in more complex and high-stakes interaction settings, supporting transparency, trustworthiness, and performance under elevated cognitive and emotional demands.From a broader perspective, this work advances affective and cognitive robotics by introducing a mechanism that bridges the gap between reactive emotional systems and reflective, self-regulating architectures.Traditional models often rely on surfacelevel mappings between external events and emotional displays, resulting in behaviours that may appear expressive but lack the deeper coherence that characterises human affective life.By incorporating inner speech, the models developed in this dissertation enable robots to engage in self-referential processing, thereby enriching their emotional landscape and enhancing their social presence.This integration opens new avenues for research on how artificial agents can develop and maintain internal narratives, how

Links and sources

• PDF from original source

Comments