Visual perception provides what was suggested to be the most important information to robots, allowing them to achieve successful interaction with human partners [94]. This information can be used in a variety of tasks, such as navigation, obstacle avoidance, detection, understanding, and manipulation of objects, and assigning meanings to a visual configuration of a scene [95,96]. More specifically, the vision has been used for the estimation of the 3D position and orientation of a user in an environment [97], the estimation of distances between a robot and users [98], tracking human targets and obtaining their poses [83], understanding human behavior aiming to contribute to the cohabitation between assistive robots and humans [99]. Similarly, the vision has been used in a variety of other applications, such as recognizing patterns and figures in exercises in a teaching assistance context in a high school [29], detecting and classifying waste material as a child would do [19], and detecting people entering a building for a possible interaction [17]. Moreover, the vision has been used in [71] for medication sorting, taking into account pill types and numbers, in [100] for sign recognition in a sign tutoring task with deaf or hard of hearing children, and in [101] as part of a platform used for cognitive stimulation in elderly users with mild cognitive impairments.

Although some applications of social robotics involve robots taking vocal commands without generating a vocal reply [102], interactions can be made richer when the robot can engage in conversations. A typical social robot with autonomous conversation ability must have the capacity to acquire sound signals, process them to recognize the speech, recognize the whole sequence of words pronounced by the human interlocutor, formulate an appropriate reply, and synthesize the sound signal corresponding to the reply, then emit this signal using a loudspeaker. The core component of this ability is the recognition of word sequences and the generation of reply sequences [103]. This can rely on a learning stage where the system acquires the experience of answering word sequences by observing a certain number of conversations that are mainly between humans. Techniques used in this area involve word and character embeddings, and learning through recurrent neural network (RNN) architectures, long short-term memory networks (LSTM), and gated recurrent units (GRU) [103,104]. It is to note that not all social robotic systems with conversational capacities have the same levels of complexity as some use limited vocabularies in their verbal dialogues. In this context, Conversation scenarios were seen in [31], verbal dialogue in [20], dialogues between children and a robot in [33], and some word utterances in [19]. A review of conversational systems usages in psychiatry was made in [105]. It covered different aspects such as therapy bots, avatars, and intelligent animal-like robots. Additionally, an algorithm for dialogue management has been proposed in [106] for social robots and conversational agents. It is aimed at ensuring a rich and interesting conversation with users. Furthermore, robot rejection of human commands has been addressed in [107] with aspects such as how rejections can be phrased by the robot. GPT-3 [108] has emerged as a language model with potential applications in conversational systems and social robotics [109]. However, in several conversational systems, problems have been reported, such as hallucinations [110,111], response blandness, and incoherence [103]. The research work presented in [87] aimed at improving the conversational capabilities of a social robot by reducing the possibility of problems as described above, and improving the human-robot interaction with an expressive face. It intended to have a 3-D printed animatronic robotics head with an eye mechanism, a jaw mechanism, and a head mechanism. The three mechanisms are designed to be driven by servo motors to actuate the head synchronously with the audio output. The robotics head design is optimized to fit microphones, cameras, and speakers. The robotics head is envisioned to meet students and visitors in a university. To ensure the appropriateness of the interactions, several stages will be included in the control framework of the robot and a database of human–human conversations will be built upon for the machine learning of the system. This database will be built in the aim of training the conversational system in contexts similar to its contexts of usage. This will increase the adequacy of the conversational system’s parameters with respect to the tasks it is required to do, and increase the coherence and consistency of the utterances it produces. For that, the recorded data will comply with the following specifications:

• Context of the interactions: Visitors approach the receptionist and engage in conversations in English. Both questions and answers will be included in the database.
• Audio recordings of the conversations: a text by speech recognition modules is used to transcript the conversation.
• Video recordings of the interaction, showing the face and upper body of the receptionist, with a quality of images usable by body posture recognition systems.
• The collected data will be used to progressively train the system. Each conversation will be labeled with the corresponding date, time and interaction parties.
• Participants will be asked to be free to ask questions they may have to inquire about the center in English, without having any other constraint or any specific text to pronounce.

Aside from the ability to process and generate sequences of words, a social robot requires more capacities to increase engagement and realism in the interaction with a human. This can be done through speech-accompanying gestures and facial expressions. Indeed, facial expression has an important role in communication between humans because it is rich in information, together with gestures and sound [112,113,114]. This issue has been studied in psychology, and research indicates that there are six main emotions associated with distinctive facial expressions [115]. At Columbia University [116], scientists and engineers developed a robot that can raise eyebrows, smile, and have forehead wrinkles similar to humans. This robot can express the face more accurately compared to the rest of the robots. This robot, called Eva, can mimic head movements and facial expressions. In this robot, 25 muscles are used, and 12 of them are dedicated specifically to the face. These muscles can produce facial skin excitations of up to 15 mm. In other works, different examples can be found for applications of gestures and expressions in social robotics. For instance, gestures have been combined with verbal dialogue and screen display in [20] for health data acquisition in hospitals with Pepper. In [40], a robot with the ability to display facial expressions was used in studies related to storytelling robots. These studies focused on the roles of the emotional facial display, contextual head movements, and voice acting. In [113], a framework for generating robot behaviors using speech, gestures, and facial expressions was proposed, to improve the expressiveness of a robot in interaction with humans. of interaction with human users.

2. Metrics of Human Perception and Acceptability