HRI

Select one option from the following list: Robots don't use computers. HRI uses only one computer, whereas HCI uses many computers. HRI focuses on embodied social agents, whereas HCI focuses on interactions with computers. HCI focuses on computers, whereas HRI focuses on humans. HRI focuses on the interaction between machines, whereas HCI focuses on the interaction between humans.

What makes HRI unique?. Embodiment makes interactions different from other computing technologies. All answer are correct. The interaction usually includes physically embodied robots. The interaction of humans with social robots.

The robot embodiment determines... Both SW and HW. The interactions with people that are affordable. All answers are correct. The physical constraints on the ways in which it can sense and act in the world.

Select the false statement. Affordances signal ways of engaging with an object but not with a robot. Affordances were introduced by J.J.Gibson, a psychologist. Affordances are perceivables relationships between an organism and its environment that enable certain actions. Objects are perceived to afford certain actions that depend on the form of the perceiver.

Select the statement that is false. Zoomorphic robots have an animal shape. Form determinies affordances. There is a discussion about it form determines function or viceversa. Humanoids have a human-like shape, androids don't.

Select the statement that is false. Engineers usually follow inside-out design approaches, like in Roomba. Designers usually follow the outside-in design approach, i.e. starting with a clay prototype. The "Frankenstein approach" consist on assembling technical components and worry about appearance and interaction later. The "Frankenstein approach" is outside-in design approach.

Select the statement that is false. Interaction expands function is a design principle. Design principles consider design patterns and design affordances disjointly. Underpromise and overdeliver are a design principles. Do not mix metaphors is a design principle.

Select the statement that is true. Anthropomorphization is the attribution of human traits, emotions, or intentions to animated entities. Anthropomorphization is an important design affordance in HRI: exploited so that humans see robots as social agents. Pareidolia is seeing humanlike features in paranormal patterns.

Select the false statement. A pattern is a problem which occurs repeteadly in our evironment, and then describes the solution to that problem, in such a way that you can use it repeteadly, without ever doing it the same way twice. A pattern is a problem which occurs over and over again in our environment, and then describes the core of the solution to that problem, in such a way that you can use the solution a million times over, without ever doing it the same way twice. A pattern is a problem which occurs repeteadly in our environment, and then describes the core of the solution to that problem, in such a way that you can use the solution a million times over, doing it in the same way.

Select the statement that is true. The uncanny valley theory explains that "the more non-humanlike a robot is, the more likable it is, until it becomes almost exactly humalike, in which case it become unlikable". The uncanny valley theory explains that "the more human-like a robot is, the more unlikable it is, until it becomes non-humanlike, in which case it becomes more likable". The uncanny valley theory explains that "the more humanlike a robot is, the more likable it is, until it becomes almost exactly humalike, in which case it become unlikable".

Select the false statement. For social scientists, anthropomorphism is attributed by humans to robots. For robot designers, anthropomorphism is a characteristic of the robot. Anthropomorphism is independent of the relationship between robot design and people's perceptions of robots. Anthropomorphism depends on the relationship between robot design and people's perceptions of robots.

Select which statement is false regarding proxemics. Proxemics researches how spatial positioning influences: attitudes and behaviurs but not interpersonal interaction. Proxemics researches how spatial positioning influences: attitudes, behaviours and interpersonal interaction. Proxemics study how people take up space in relation to others.

Adam Kendon (1990) defined F-formations which are defined on three spaces. Can you link each concept to its corresponding definition?. The space where the participants find themselves. The inner space between the people involved in an interaction. The surronding space.

Choose the statement that is false. People from different cultures have varying personal proxemic preferences and expectations. Proxemics are not relevant in human-drone interaction since drones fly. When placement a robot in space, it is crucial to consider people's preferences and the social norms that exist in relation to others. Robot behaviour that can be considered rude/innapropiate must be avoided.

Make the statement that is false. Researchers found that people sitting next to each other behave more compeitively, whereas people sitting opposite to each other behave more cooperatively. Proxemic norms are sometimes broken in order to intimidate. The meaning of spatial-interaction cues is highly contextual. Proxemic norms are sometimes broken in order to create more psychological closeness.

Match the corresponding definitions to the correct concept. Construct a map while the robot moves and localizes. Move from waypoint to waypoint avoiding obstacles. Determinate position by sensing rotation of wheels. Determing a mobile robot's position and possibly its orientation. Moving to a goal location through a possibly corwded environment and maintain social distance.

Hall et al. (1968) described four distance zones in their original work: intiate distance, personal distance, social distance and public distance. Select the orientative thresholds limiting these distances when the space is (relatively) unlimited. Social space. Private space. Public space. Intimate space.

Choose the statement that is false. A robot that displays coherent nonverbal behaviour will make the interaction seem more natural and smooth. To develop culturally competent care robots might not involve developin cultural knowledge representations, culturally sensitive planning and execution, and culturally approapriate multimodal HRI. When speaking to a robot, we would expect that the robot turns its head toward us.

Select the statement that is false. Gestures can be used for emphasizing particular moments during speech. Gesture generation SW is trained by feeding it hundreds of hours of video recording of people talking, allowing machine learnig to build a model that generates gestures that closely fit spoken communication. Deictic gestures involve pointing to specific things in the enviroment which help disambiguating but it does not help to establish joint attention. The generation of co-speech gestures for artificial agents is now largely done by machine learning.

Choose the statement that is false. Joint attention refers to the interaction partners attending to the same areas or objects at different times. In a multiparty interaction, a robot can use its gaze to control who will be the next person to talk. Joint attention with robots has been studied in interactions with children who have autism, with the aim of using the robot to assit them in developing this important social skill. Researchers have used human gaze behaviour to derive models of gaze behaviour for robots.

Select the statement that is false. Imitation is the conscious replication of another's behaviour. Mimicry is central in human cognition and it has been found to have a basis on mirror neurons. Learning by imitation is a technique in robitcs where robots record human actions and then they reproduce these actions. Mimicry is the unconcious replication of the behaviour of another person. Mimicry, as a largely automatic behavioural response, has many significal social functions; one is that indirectly signals negative affect and disliking for an interaction partner.

Classify the following statements in true or false. Gaze is only important for showing someone that you're listening to them. Nonverbal behaviour is not very relevant for human-robot interactions. Nonverbal behaviour is mostly conscious. Proxemics (i.e., social distance) is a part of nonverbal communication. Nonverbal and verbal behavior always align. Nonverbal behavior can be used to predict social outcomes, such as whether a couple will stay together. Mimicry is an unconscious behavior, and imitation is a conscious behavior.

Select the statement that is false. An interaction can appear awkward when the robot produces gestures that do not match the rhythm or meaning of its speech. Eye gaze signals shared attention and it nevers occurs automatically. Nonverbal cues allow people to communicate information "between the lines". Keepon robot uses gaze and reactive motion to express attention and affect. In the HRI context, verbal and nonverbal cues might be contraditory.

Select the statement that is false. A robot's posture can be used to express motion, and through that, affect the emotion of onlookers. Posture and gesture recognition are achieved using standard pattern-recognition techniques that can be used in robots to perceive and identify human nonverbal cures. In verbal communication, timing refers to turn-taking among interaction partenrs. Nonverbal cues do not need to be contingent on the interaction.

Select which statement is false. Touch does not always have positive effect in HRI. Physical interaction with companion robots like Paro baby seal robot is uncommon due to lack of safety haptic interactions. Physical interaction with humanoid robots is relatively uncommon for a number of reasons, e.g. lack of safety during simple actions like hugging or shaking hands. Touch is nonverbal cue that is often involved in close interactions among people and it is influenced by the social context.

Select the statement that is false. In Space Odyssey, HAL does not open the pod bay door because it found out a written message reading that the astronauts were going to disconnect it. Language understanding in HRI can do intent recognition by indentifying keywords (e.g. command, location, person, event, date, etc.). A robot that was programmed to take everything literally was confused when the commands were given as questions, that is, it was not interpreting politeness expressions (according to the studies by Williams et al, 2018).

Select the statement that is false. Extracting semantic content, that is, what the utterance means is difficult. Sentiment analysis is used to extract the affect of an utterance. Sentiment analysis is optimized for written language, but not for spoken language. SW tools that analyses a message can obtain a measure that says how positive or negative is in percentages (e.g. 6% angry, 87% happy, 9% excited). Prosody (rhytm, beat, ...) and amplitude help to understand if the speaker is happy or agitated (without knowing the language).

Select the statement that is false regarding Large Language Models (LLMs). LLMS relying on new ML technology (e.g. transformers) with billions of parameters. LLMs are trained using large corpora of text (e.g. Wikipedia). Language models are an AI technique built to predict an unknown word based on the words around it (or even the words just preceding the unknown word). In seconds, LLMs can write prose; compose poetry; and respond to questionsabout the population and size of New Zealand, quantum computing, etc. LLMs can evaluate if the text they produce is correct or incomplete.

Select the sentence that is false. Phonemes are small units of sound that make up words. Words are the largest units of sound we can utter to convey meaning. Speech comprises utterances (the smallest unit in spoken language).

Select the statement that is false. Up until 2010, speech recognition engines relied oun rule sets extracted from the analysis of speech data. They often used Gaussian Mixture models (GMM) and Hidden Markov models (HMM). The models generate a probabilistic model of how phonemes are strung togeher to form a valid sentence. The state-of-the-art techniques can deal with background noise and crowded environments, but they are gender dependent. The state-of-the-art techniques for speech recognition involve deep-learning approaches which implictly perform a similar process to GMM and HMM using large-scale neural networks ans sequence to sequence models. The Gaussian mixture models and the Hidden Markov models were the best solution for speech recognition for decades.

Inidicate which of these statements is false. The performance of current speech recognition techniques is better than that of human transcribers. Speech recognition engines are cloud-based services nowadays. A voice recording is streamed to a server where powerful computers with hardware acceleration transcribe the recording almost instantly. Speech recognition techniques cannot run on board the robot.

Select the statement that is false regarding "Speech Recognition": Speech recognition techniques understand or interpret what has been said. Speech recognition is also known as Speech To Text (STT). Speech recognition is also known as Automated Speech Recognition (ASR). It requires recording with a high-quality microphone and positioning close to the speaker.

Select the statement that is false. Speech-recognition engines generally have a simplee-to-use application programming interface (API), allowing the programmer to quickly integrate speech recognition into the robot. Voice-activity detection is often part of ASR and it distinguishes speech from silence but it cannot distinguish music playing from someone talking. If cloud-based ASR cannot be used, then onboard speech-recognition solutions that use a reduced neural network or a classical approach to speech recognition can be used. ASR engines often return a confidence value next to the trascribed sentence, giving a measure of how confident the engine is about the transcribed speed.

What is the smallest unit that a speech recognition engine tries to recognize? Select one option from the following list below: Homophone. Phoneme. Synomym. Word. Letter.

Select the false statement regarding LLMs. In HRI, most applications require fluent and interactive two-way conversation between a person and a robot (dialogue management and chatbots) which is fully achieved by LLMs. LLMs, such as BERT or GPT, are trained on huge data sets of text. The training in transfer learning uses a smaller set of training data usually to do something domain specific (e.g. such as predicting the next word in book reviews). If we use sentences of two kinds i.e. help and complain then we can apply a fine-tuning step and the model will be able to classify the intent of new sentences into these 2 categories. Transfer learning is typical in computer vision: e.g. a neural network trained on a huge dataset to classify images, then it can be repurposed to detect a specific thing (e.g. a bird) using transfer learning.

Select the statement that is false regarding speech recognition. The speech is converted from the time domain to the frequency domain because this makes phonemes recognition easier. Speech is recorded digitally in the time domain. Phoneme have different characteristics that are easier to recognise in the time domain. In order to record speech, the amplitude of the signal is recorded every 1/16000 fraction of a second.

Select the statement that is false. Verbal communication is enriched by paralinguistic information such as rythm, timing, gaze, gestures and facial expressions. Speech is not open to multiple interpreations. Producing robot speech is much simpler than understanding human speech, which creates an imbalance between people's expectations and robot's actual capabilities. Intonation and emphasis can change the meaning of a sentence substantially.

Select which statement is false. All speech-recognition engines still struggle with recognizing atypical speech. Noisy, reverberating, or crowded spaces will decrease ASR performance. Local dialects of languages or nonnative speakers could result in reduced recognition performance. Proper nouns, such as Margaret or Launceston Street, are easily picked up correctly than common nouns.

Select the statement that is false regarding word embeddings. word2vec embeddings convert words or text to a vector (a series of numbers). Analogies can be computationally solved by substracting word embeddings. Sarcasm can be computationally solve by dividing and multiplying word embeddings. Words that are similar in meaning are described by vectors that are closer (evaluated by a distance measure: the cosine similarity). Neural networks learned associations between words (relations or links) by analysing the co-ocurrence of words and phrases in large amounts of text. These vectors can ecode the semantic and syntactic relations between words.

Select the statement that is false. Voice-activity detection is used to give the impression that the robot is listening. A common misconception is that Automated Speech Recognition (ASR) also means that the speech is understood. Sarcasm is not still a challange in AI.

Select the statement that is false regarding chatbots that are powered by LLMs. Current research is looking into using LLMs to facilitate a goal-oriented interaction. There is a friendly back-and-forth between the robot and the user. The Chinese room effect does not take place. Conversations tends to go away from the topic and divagate/wander without aim.

Select the false statement regarding the use of verbal interaction tools in HRI. Current HRI developers can use online services for: STT and TTS. Chatbots and LLMs have been used to build open-ended spoken HRIs. OpenAI offers GPT technology, Google offers Cloud Speech API, Microsoft has its Azure Cognitive Services, and Amazon offers its Alexa set of tools. For speech-synthesis: The robot sends what it wants to say to a server and it receives back the audio signal and the robot palys that on its loudspeakers. For speech recognition, an audio signal is recorded through the robot's microphone and send to the servers, there is analysed in real-time and tehy send back teh recognized text while the user is still talking. However, these services cannot indentify entities, syntax, sentiments, and categories in the talking.

Select the sentence that is false regarding Dialogue Managers (DMs). In User-initiative DMs, the user takes the lead and the DM intervenes when more information is needed. These DMs use a planner. Simple DMs are composed by a set of rules that define different states of the system (i.e. the dialogue) and conditions governing the change from one state to another, that is there is a strict control of flow. Simple DMs are finite state machines (FSMs). Modern DMs use the power of LLM transformers. In User-initiative DMs the flow is linear and the dialogue cannot be interrupted.

There has been recent growth in technologies to create chatbots. Which of the following statements is false? Select one or more options from the following list: A chatbot is sometimes created based on more than several terabytes of data. There is a chatbot that can explain what an HRI is. There are only chatbots that can answer questions for a limited amount of topics. There is a chatbot that can write a simple program, such as a sorting task (it is a simple programming task often used for beginners's practice).

There are two approaches for generating artificial speech introduced in this chapter: parametric TTS and neural vocoder. Which of the following statements are true? Select one or more options from the following list: Neural vocoder is a deep-learning-based method trained on a large amount of speech data. Parametric TTS produces speech that is virtually indistinguishable from human speech. Neural vocoder is a deep-learning-based method trained only on a large amount of text data. Parametric TTS is better than neural vocoder in the aspect that neural vocoder can only be implemented as a cloud-based service.

Select which statement is false regarding Dialogue management (DM). DM keeps track of the state of the conversation: - What the robot needs to know (e.g. the order and the extras) - What the robot knows (e.g. the order: pizza) - What the robot has yet to establish (e.g. pepperoni). QiChat is the DM software used for Softbank Robotics robots, such as the Nao and Pepper robots. System initiative DMs are suitable for closed and well-contextualized tasks such as registering guests, filling out forms, taking orders, etc. A command (e.g. "play Bob Marley") does need DM, but a request (e.g. "can I have a drink?") does not. Dialogue management (DM) is the process that keeps a conversation flowing.

Select which statement is false regarding chatbots. Chatbots such as Siri (Apple), Cortana (Microsoft), Alexa (Amazon), and Bixby (Samsung) respond to simple spoken commands and, if needed, respond with spoken language. Combining chatbots with robots still have challanges since robots are multimodal. Chatbots are computer programs intended to maintain a conversation with the user, typically through using text. Bots now remember the context: your name, where you went on holiday and what you were so excited about at the starting of the chat, but they forget all this information when the chat finishes. Genuine understanding of natural language, in a way that we would consider to be humanlike, is finally achieved by LLMs. Chatbots are also general-purpose agents that try to respond to unconstrained input. They are built using thousands of handcrafted rules on how to respond to often-occurring utterances or built by maintaining a database of all previous conversations, from which responses are fitted. Chatbots developed using ML can generate inappropiate reponses (e.g. Tay by Microsoft).