Outline for 4/16
Logistics
Course Topics by Week
Unifying View of AI
Specifying a search problem?
Search Summary
Binary Constraint Network
Constraint Satisfaction Summary
Backjumping (BJ)
Other Strategies
Nodes Explored
Knowledge Repr. Summary
Resolution
Davis Putnam (DPLL)
GSAT
Today’s Outline
Emergence of Complex Autonomous Systems
Mobile Robots
Immobile Robots Example 1: Autonomy on a Global Scale
Immobile Robots Example 2: Cassini Saturn Mission
Properties of Complex Autonomous Systems
NASA’s Autonomous Systems
Additional Examples of Complex Autonomous Systems
Programming Complex Autonomous Systems
Challenge: Autonomous Saturn Orbital Insertion of Cassini
Cassini Propulsion System Schematic
Reconfiguring for a failed engine
Autonomous System Coding Challenge
Solution: Part 1 Model-based Programming
Single Core Model Describes Software, Hardware Dynamics
AI Code Identifies + Reconfigures Component Modes
Traditional Control System
Solution: Part 2 Model-based Deductive Executive
Autonomy architecture with hybrid model-based & scripted executives
Combining Reactivity and Deduction: Is it practical?
Can autonomous systems be built w/ low cost, robust reactive networks?
SAT, CSP problems for physical systems are typically easy
Solution: Part 3 Risc-like Best-first, Deductive Kernel
Deep Space One
Consider a sub family of model-based optimal controllers where...
A family of increasingly powerful deductive model-based optimal controllers
Modeling the Plant
Specifying a valve
Configuration System (S, ??
Model-based Configuration Manager
Mode identification + reconfiguration
Mode identification via Consistency-based diagnosis
Example: Cassini propulsion system
More conflicts
A candidate covers each conflict
Mode Identification as CSP
Mode Reconfiguration
Reconfiguring to restore thrust
Conflicts focus search
Statistically Optimal Configuration Management
MI and MR performance
Diagnosis of Combinatorial Circuits
Combinatorial optimization problem
MI/MR as combinatorial optimization
Simple cost model
Using the simple cost model for MI
Limitations of the simple cost model
Best first search
Required subroutines for BFS
Representing assignments
Basic successor function
Successor lattice
Conflicts
Focusing with conflicts
Initializing the agenda
Assignment {} is infeasible
Assignment {v1=b1} is infeasible
Least cost feasible assignment found
Decreasing agenda size
Only {v1=b1} added to agenda
Immediate successor and sibling of {v1=b1} added to agenda
LTMS: Truth Maintenance System
Example: DS-1 bus communication
Generic LTMS interface
Using the LTMS in MI and MR
ITMS = Incremental LTMS
LTMS labels
Conflicting clauses
Unit propagation: the basic idea
Label inference + proposition support
LTMS after initialization
Propagating C10
Propagating C7
Propagating C6
Unit propagation at the fringe
Unit propagation algorithm
Updating fringe and conflicts
Well-founded support
Implementing the generic interface
Incrementally modifying ?
Before deleting C11
After deleting C11
Email: weld@cs.washington.edu
Other information: CSE 592, Lecture 3
Download presentation source