* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Architectures for Robot Control
Survey
Document related concepts
Stanford University centers and institutes wikipedia , lookup
Visual servoing wikipedia , lookup
Agent-based model in biology wikipedia , lookup
Ecological interface design wikipedia , lookup
Mathematical model wikipedia , lookup
Intelligence explosion wikipedia , lookup
Existential risk from artificial general intelligence wikipedia , lookup
Self-reconfiguring modular robot wikipedia , lookup
Embodied cognitive science wikipedia , lookup
Philosophy of artificial intelligence wikipedia , lookup
Knowledge representation and reasoning wikipedia , lookup
Transcript
Architectures for Robot Control Intelligent Robotics 2014/15 Bruno Lacerda This Lecture • Deliberative paradigm - • STRIPS Reactive paradigm - Behaviour-based architectures • Hybrid paradigm • The SMACH package (ROS) Intelligent Robotics 14/15 2 Architectures for Robot Control - Overview Deliberative Paradigms • Sense the world • Update world model and generate plan - • Knowledge representation + Automated reasoning Execute planned action Bottleneck Classical / Hierarchical Paradigm Sense Plan Act ! 70’s ! Focus on automated reasoning and knowledge representation ! STRIPS (Stanford Research Institute Problem Solver): Perfect Intelligent 14/15 world model, closed worldRobotics assumption 4 Deliberative Paradigms • Knowledge representation: - Closed world assumption: World model contains everything the robot needs to know - Frame problem: How to model everything the robot needs to know while keeping the size of the state space manageable? Classical / Hierarchical Paradigm Sense Plan Act ! 70’s ! Focus on automated reasoning and knowledge representation ! STRIPS (Stanford Research Institute Problem Solver): Perfect Intelligent 14/15 world model, closed worldRobotics assumption 5 Deliberative Paradigms • Planning: - Exploring the possible states of the world to find a plan is computationally expensive (e.g., STRIPS planning is NPcomplete) Classical / Hierarchical Paradigm Sense Plan Act ! 70’s ! Focus on automated reasoning and knowledge representation ! STRIPS (Stanford Research Institute Problem Solver): Perfect Intelligent 14/15 world model, closed worldRobotics assumption 6 Deliberative Paradigms • Sensing and acting are disconnected: - After sensing we always need to update the world model and generate a plan - No reactivity Classical / Hierarchical Paradigm Sense Plan Act ! 70’s ! Focus on automated reasoning and knowledge representation ! STRIPS (Stanford Research Institute Problem Solver): Perfect Intelligent 14/15 world model, closed worldRobotics assumption 7 Shakey • Control architecture based on deliberative paradigm • Plan step is based on STRIPS - Knowledge representation: First-order predicate logic - Planning: Logic based planning - Remember lectures on classical planning Intelligent Robotics 14/15 8 Shakey Intelligent Robotics 14/15 9 Reactive Paradigm • No world model; no planning • Direct mapping between sensor input and actuators output • Very reactive to changes in sensor readings • (Radical) answer to shortcomings of deliberative paradigm Reactive / Behavior-based Paradigm Sense Act ! No models: The world is its own, best model Intelligent Robotics 14/15 10 Reactive Paradigm “ AI researchers ... partition the problems they work on into two components. The AI component, which they solve, and the nonAI component which they don’t solve. Typically AI ’succeeds’ by defining the parts of the problem that are unsolved as not AI. The principal mechanism for this partitioning is abstraction ... In AI, abstraction is usually used to factor out all aspects of perception and motor skills. I argue below that these are the hard problems solved by intelligent systems, and further that the shape of solutions to these problems constrains greatly the correct solutions of the small pieces of intelligence which remain.” Rodney Brooks, Intelligence without representation, 1991, Artificial Intelligence, Vol 47. Intelligent Robotics 14/15 11 Behaviour-Based Architectures • • Overall controller composed of two parts - Task achieving controllers (e.g., simple state machines) - Arbitrating controller (also task specific) Controllers: - Are reactive (map perception to action) - Do not include world models, no planning - Are only capable of performing one task each ‣ Should perform it very efficiently Intelligent Robotics 14/15 12 Behaviour-Based Architectures Push Go through door Arbitration Approach object Follow a wall Intelligent Robotics 14/15 13 Behaviour-Based Architectures • Behaviours are the basic building block for robotics actions, with an overall emergent behaviour obtained from their arbitration • ︎Behaviours support good software design principles due to modularity • However, we have some assumptions: - We need to be able to decompose the task into appropriate “primitive” behaviours - Each “primitive” behaviour requires only simple perceptual and physical talents Intelligent Robotics 14/15 14 Behaviour-Based Architectures Different behaviourbased architectures have different approaches to provide arbitration Push Go through door Arbitration Approach object Follow a wall Intelligent Robotics 14/15 15 The Subsumption Architecture • Behaviours are networks of sensing and acting modules (finite state machines) • ︎ Modules are grouped into layers of competence • Top layers can subsume lower layers Intelligent Robotics 14/15 16 Herbert Intelligent Robotics 14/15 17 Shortcomings of Behaviour-Based Architectures • Very task oriented - no flexibility • No world model means no memory or ability to reason about optimal behaviours • Arbitrating a large set of behaviours properly is a cumbersome and error-prone task • Complex tasks require arbitration of a large set of behaviours Intelligent Robotics 14/15 18 Hybrid Paradigm • World model, used for planning • Closed-loop, reactive control • Many architectures of today follow this paradigm Hybrid Deliberative/reactive Paradigm Plan Sense Act ! Combines advantages of previous paradigms ! Intelligent Robotics 14/15 World model used for planning 19 Reading • Deliberative paradigm - [Fikes, Nilsson (1971)] STRIPS: a new approach to the application of theorem proving to problem solving. Artificial Intelligence, 2:189-208. - http://www.sri.com/work/timeline-innovation/timeline.php? timeline=computing-digital#!&innovation=shakey-the-robot Intelligent Robotics 14/15 20 Reading • Reactive paradigm - [R. A. Brooks] ‣ (1986), ”A Robust Layer Control System for a Mobile Robot", IEEE Journal of Robotics and Automation RA-2, 14-23. ‣ (1987), "Planning is just a way of avoiding figuring out what to do next", Technical report, MIT Artificial Intelligence Laboratory. ‣ (1991), "Intelligence Without Representation", Artificial Intelligence 47 (1991) 139-15 Intelligent Robotics 14/15 21 Reading • General bibliography: - Behavior-based Robotics [Arkin] - Introduction to AI Robotics [Murphy] Intelligent Robotics 14/15 22 The SMACH package Overview • Python implementation of robot executive based on hierarchical state machines • Also provides concurrency • See http://wiki.ros.org/smach Intelligent Robotics 14/15 24