180 likes | 275 Vues
Explore the dynamic world of a Lisp program agent operating in a simulated laundry environment. Learn about RAMA architecture combining plan execution and process monitoring for achieving goals and adapting to changes dynamically. Discover the role of expectations, transitions, interpreter, and adaptation in this autonomous agent system. See how intelligent agents sense, act, and learn in pursuit of complex activities, with a focus on handling unexpected events and failures effectively.
E N D
What am I? • Lisp Program that can operate equipment in the virtual laundry world.
Why Am I an Agent? • I live in environment (simulated) • I sense my environment • I act on it (wash clothes) • Over time • In pursuit of goals (clean clothes) • My actions affect my future senses • I AM AN AGENT!!!!!!!!!
What is my Control Structure? • RAMA – Routine Activity Management and Analysis • RAMA is an autonomous agent architecture which combines plan execution and process monitoring. • RAMA uses the same knowledge representation for plans and processes (expectations)
Dynamic • “Regularities in the interaction between certain kinds of agents and certain kinds of worlds” (Agre 1989) • Agent is both observer and actor. Therefore, dynamic merges the notion of plan and process: dynamic contains actions and events to expect while involved in the activity
Current World Model contains assertions about the objects participating in the active dynamics. Assertion has 4 elements: {property, object, value of the property, time point}
Expectations • Temporal Queries to the Current World Model • Query has 4 arguments: property, object, value for the property, time point. The queries are EQUAL, >,< • Time point can be a number or a range
Transitions • T has five elements {Q1, Q2, d, sigma, D) • Q1, Q2 are queries • d is the mean time it took from the time Q1 was true till the time Q2 was true • Sigma is standard deviation • D is a vector of last N instances of T • T is a prediction of time duration
Interval Based Queries Used to define transition based expectations
Interpreter • Match queries in Active Expectations against assertions in Current World Model to confirm or detect failed Expectations • Generate expectations • Select actions.
Dynamic is defined using • Events • Interval Clauses • Actions • Goals • Process Combinators: Sequence, Parallel, Cycle, Either, Suspend, Stop
Uniform representation – in LISP everything is a node Action Dynamic Transition Event Expectation
Adaptation • Update run-time duration estimates • Handling Unpredicted Assertions • Handle Expectation Failures
Final Remarks • Autonomous Agent Architecture implemented in LIST and successfully performed in the simulated environment • Generating and verifying expectations is an essential component of performing complex purposeful activity in structured environments. • Learning using human assistance