AI and Robotics: Lessons Learned from the RACE Project · AI and Robotics: Lessons Learned from the RACE ... Ontological models: ... AI and Robotics: Lessons Learned from the RACE
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
AI and Robotics: LessonsLearned from the RACEProject
A. Saffiotti, T. Cohn, K. Dubba, J. Hertzberg,L. Hotz, Š. Konecný, J. Lehamnn,L. Lopes, M. Mansouri, B. Neumann,M. de Oliveira, S. Rockel, S. Stock, L. Zhang
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Model-Centered Approach
“Robotics is closely related to AI [as] intelligence isrequired for manipulation, navigation, localization,mapping, motion planning”[Wikipedia page on Artificial Intelligence]
Model-centered approach is a major contribution of AI toRoboticsModels capture environment, capabilities, tasks
“competent behavior” results from reasoningmodels have formal propertiesmodels can be changed to suit different environments,physical capabilities, tasks
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Model-Centered Approach
“Robotics is closely related to AI [as] intelligence isrequired for manipulation, navigation, localization,mapping, motion planning”[Wikipedia page on Artificial Intelligence]
Model-centered approach is a major contribution of AI toRoboticsModels capture environment, capabilities, tasks
“competent behavior” results from reasoningmodels have formal propertiesmodels can be changed to suit different environments,physical capabilities, tasks
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Model-Centered Approach
“Robotics is closely related to AI [as] intelligence isrequired for manipulation, navigation, localization,mapping, motion planning”[Wikipedia page on Artificial Intelligence]
Model-centered approach is a major contribution of AI toRoboticsModels capture environment, capabilities, tasks
“competent behavior” results from reasoningmodels have formal propertiesmodels can be changed to suit different environments,physical capabilities, tasks
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Scenario: a Busy Restaurant
knife
fork
dish
time
Spatial models: cappuccinosshould be served “in front of”guests
Resource models: cappuccinoswon’t fit if other dishes arepresentAction models: dishes can becleared to make space forcappuccinosOntological models: dishes andcutlery should be cleared, saltand pepper should stayTemporal models: cappuccinosshould be served before they getcold
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Scenario: a Busy Restaurant
knife
fork
dish
time
Spatial models: cappuccinosshould be served “in front of”guestsResource models: cappuccinoswon’t fit if other dishes arepresent
Action models: dishes can becleared to make space forcappuccinosOntological models: dishes andcutlery should be cleared, saltand pepper should stayTemporal models: cappuccinosshould be served before they getcold
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Scenario: a Busy Restaurant
knife
fork
dish
time
Spatial models: cappuccinosshould be served “in front of”guestsResource models: cappuccinoswon’t fit if other dishes arepresentAction models: dishes can becleared to make space forcappuccinos
Ontological models: dishes andcutlery should be cleared, saltand pepper should stayTemporal models: cappuccinosshould be served before they getcold
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Scenario: a Busy Restaurant
knife
fork
dish
time
Spatial models: cappuccinosshould be served “in front of”guestsResource models: cappuccinoswon’t fit if other dishes arepresentAction models: dishes can becleared to make space forcappuccinosOntological models: dishes andcutlery should be cleared, saltand pepper should stay
Temporal models: cappuccinosshould be served before they getcold
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Scenario: a Busy Restaurant
knife
fork
dish
time
Spatial models: cappuccinosshould be served “in front of”guestsResource models: cappuccinoswon’t fit if other dishes arepresentAction models: dishes can becleared to make space forcappuccinosOntological models: dishes andcutlery should be cleared, saltand pepper should stayTemporal models: cappuccinosshould be served before they getcold
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Key Challenges
? What modeling languages areappropriate?
How do we jointly reason aboutdifferent models?How do we learn relevant modelsfrom few examples?How to we obtain symbolic models(appropriate for reasoning) fromexperience?
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Key Challenges
? What modeling languages areappropriate?How do we jointly reason aboutdifferent models?How do we learn relevant modelsfrom few examples?How to we obtain symbolic models(appropriate for reasoning) fromexperience?
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Conceptualization, Generalization and Unseen States[L. Hotz, B. Neumann, S. von Riegen, N. Worch. “Using Ontology-based Experiences for Supporting Robots Tasks”]
How is AI Useful for Robots? Knowledge Representation Learning from Few Examples Planning & Execution Conclusions
Using the Most Appropriate KR&R Approach
Not all reasoning tasks fall within the scope of OWL-2-DLnot all types of knowledge are easy to represent inOWL-2-DLnot all reasoning tasks are amenable to DL inference
Reified constraints
Class: SceneLayoutEquivalentTo: Occurrence
AND (hasPassiveObject SOME PassiveObject)AND (hasLayoutConstraint EXACTLY 1 LayoutConstraint)