Ashita Ashok

• Highly anthropomorphic social robots create a difficult interaction problem because their bodies suggest social, perceptual, and cognitive capability before the implemented system has shown what it can reliably support. This thesis examines that problem through Ameca, a full-body humanoid robot whose face, gaze, voice, speech, upper-body movement, and public recognisability make system behaviour socially interpretable. The work develops the 5A framework as an organising analytical lens for the anthropomorphic expectation gap. The framework diagnoses expectation mismatch through Anthropomorphism, Apparent Ability, and Anxiety, and addresses it through Alignment and Awareness. Across small-scale empirical studies and system evaluations with N = 929 reported participants, the thesis examines Ameca and the developed Empathic Mechanized Anthropomorphic Humanoid System (EMAH system) across embodiment, auditory dialogue, visual perception, behaviour generation, and tandem-language practice. The embodiment studies show how identity, voice, gender, gaze, affect, movement, and failure can be interpreted as cues from one social agent. The auditory-dialogue work connects speech recognition, Large Language Model (LLM)-based response generation, memory, retrieval, prompt boundaries, and lip-synchronised Text-to-Speech (TTS) to the stability of Ameca’s robot identity. The visual-perception work shows how eyes create expectations of seeing and fair attention, making Visual Question Answering (VQA), gaze control, active-speaker detection, and Hero selection socially consequential. The behaviour-generation work shows how facial expression, gaze shifts, mimicry, and pointing make ability visible, while timing, actuator limits, gesture detection, and morphology constrain what the robot can express or indicate. The tandem-language studies apply the framework in a bounded use case, where Ameca supports low-pressure German speaking practice through controlled dialogue, short turns, supportive correction, and clear role limits. The central contribution is an implementation-grounded account of expectation management for highly humanlike robots. The thesis shows that usability and acceptance depend on the fit between apparent ability, actual capability, interaction role, and failure behaviour. For researchers and designers, the findings indicate that appearance, sensing, dialogue, memory, movement, prompts, explanation, and repair need to be designed as one interaction system. Ameca can support meaningful Human-Robot Interaction (HRI) when interlocutors can recognise both the social invitation of the body and the technical boundary behind it.