2nd Test Flashcards
Properties of Motion
Safety: motion should be safe and not harm humans
Functionality & Eficiency : the motion should allow the task to be perform, taking into account that the human is acting in the world and optimizing the motion for task at hand
Predictability & Legibility: the motion should be what the user expects. The collaboratory should be able to recognize the intent of the robot
Believability: the motion gives illusion of live
Social & Culturally adjusted : the motion should respect the social and cultural norms of human-human interaction
Types of motion
Functional motion: motion that reaches its goal, avoiding collisions
Predictable motion: motion performed by the robot that is expected
Legible motion: functional motion in which the collaborator can quickly and confidently infer the goal
Disney animation principles
Squash and stretch : hard due to the robots rigid parts
Anticipation: allows the viewer to understand what the character is going to do ; allows the viewer to interpret the character in a more natural way
Secondary actions: aids in making it more believable
Arcs: states that movement should be done in arcs to look more natural
Exaggeration: enphasize the robots movements and expressions to make it more convincing
Goal oriented skill vs mean oriented skill
Goal oriented skill - related with achieving a particular world state
Mean oriented skill- include a gesture or communicative intent
What are the emotional social robots ?
Emotional Social Robots are social robots that respond emotionally to certain situations
What is the James Lange Theory
States that emotions are a result of a physiological response to stimuli
Verbal Communication systems- definitions
Dialogue modelling - formal characterization of dialogue, envolving context, and possibly continuations
Dialogue system - system that engages in dialogue
Dialogue manager - module of the system concerned with managing dialogue and decisions on how to contribute to a conversation
KnowDiaL
input
parser
grounding model : gets information from KB on hoe to infer the correct action upon a taken command
knowledge base : stores groundings of previous dialogues
dialogue manager - asks openeval to search on World Wide Web the missing parameters from the grounding model
openeval
What are the components of dialogue system
Perceive and Understand
-> NL system : generator using developed grammar for human robot interaction
-> Speech recognizer
Generate and Shynthesize:
-> NL system
-> TTS : speech shynthesizer for robot speech
Manage dialogue and Non verbal behaviours
-> Dialogue Manager
-> Gestures and non-verbal behaviour handler
What are the mechanisms for managing conversations
Role signaling mechanism: interlocutors of conversations engage in a discourse at different levels of involvement
Turn taking: role shifts from participants through a turn taking mechanism
Topic Signaling mechanism: the speakers create a discourse, which is a combination of discourse segments in structures. Speakers create a number of cues (verbal and non verbal) that signal these structures and allows the contribution of other participants
How does interaction analysis work?
Using objects affordances to anticipate the humans next move, allowing the robot to plan ahead for a more reactive response
Using simulation theory for intention recognition. Simulation theory states that people atribute mental states using their own mental processes
Learning from demonstrations
Key idea: maximize the generalization of a learnt skill to unseen situations using a minimal number of demonstrations provided by the teacher
Mapping Function: Demonstration data is used to approximate a function mapping the robots state observations to actions
System model: Demonstration data is use dto determine a possdible reward function. Policies are later derived from this info
Plans : Demonstrations data is used to learn rules that associate a set of pre and post conditions to each actions and possibly a sparsified state dynamic model
What is the correspondence problem
Refers to the problem of indentification of a mapping that allows the transfer of information from the teacher to the learner.
Record Mapping vs Embodiment Mapping
Record Mapping referes to whether the exact actions experiencied by the teacher during the demonstration execution are recording within the data set
Embodiment Mapping refers to whether the actions recorded within the data set would be exactly those that the learner would learn
Record Mapping vs Embodiment Mapping
Record Mapping referes to whether the exact actions experiencied by the teacher during the demonstration execution are recording within the data set
Embodiment Mapping refers to whether the actions recorded within the data set would be exactly those that the learner would learn
Active learning
allows for one to have more control over what examples is given
Queries
A Querie is when we choose an unlabed instance and ask for a label or when we ask for an action in a choosen state
Label queries: robot performs a said actions and inquires if it was done correctly
Demonstration queries: the robot asks for a demonstration from the user
Feature queries: involves asking if a particular feature is relevant for the learnt skill
What are the trust related characteristics?
Competence - the trustee competence in performing an action effectively
Predictability - the trustee competence in performing an action as it is expected
Benevolence - the trustee intrinsic and positive intentions towards the trustor
Integrity- the trustee’s adherence to a set of principles accepted by the trustor
Trust vs Trustworthiness
trust is in the relation
trustworthiness is a property of the trustee
Disposition to trust a robot
the extend to which the user displays a consistent disposition to willingly depend on the robot in certain situations
Situation-based trust
when the user believes the needed conditions are in place for him to predict a sucessful outcome by the robots actions
Types of gaze
mutual gaze : the gaze of each individual is in one another
gaze following: a understands that b’s gaze is not on them and follows b’s line of sight on to a point in space
joint attention: similar to gaze following but there is a focus of attention
shared attention: composition of gaze following and joint attention, the individuals gaze is in each other and in an object
theory of mind: composition of the above and uses higher order cognitive processes to infer about the other attention
Taxonomy of gaze
mutual gaze: eye gaze of one individual is on the others eyes or face
referential attention: gaze is in a location or point in space
joint attention
adverting gaze: shifts of gaze away from the main direction of the gaze
Types of nonverbal behaviour
Emblems: gestures that can be directly translated into words and have different meanings across cultures
Illustrators/Iconic Gestures: gestures or expressions that accompany speech to make it more vivid
Regulators: non verbal behaviours that regulate conversation
Self-adaptor: unconscious behaviours that release nervous energy
display of emotions
Automated Face Analysis
Input
Facial landmark detection : tracks a dense set of facial featurs
Face alignment: Removes the effects of spatial variations ; obtains canonical shape and orientation; aligns shape model with unseen images
Feature extraction: composition of landmarks and relation between them
Dimensionality reduction
Action Unit Classification
-> Static modeling: each videoframe is evaluated independently
-> Temporal modeling: the frames are segmented in to sequences and modelled with a dynamic bayesian model
