Brian C. Williams 16.410/16.413 September 3 rd , 2003 Introduction to Principles of Autonomy and Decision Making 1 Brian C. Williams 16.410/16.413 September 3 rd , 2003 Today’s Assignment Today’s Assignment • R ead Chapters 1 and 2 of AIMA ead Chapters 1 and 2 of AIMA – “ Artificia Artificia l Intelligenc l Intelligence : A Modern Approach” : A Modern Approach” by Stuart Russell and Peter Norvig by Stuart Russell and Peter Norvig –2 nd Edition (not 1 st Edition!!) –2 nd Edition (not 1 st Edition!!) – AIMA is available at the Coop AIMA is available at the Coop 1
22
Embed
l1 intro 9 rpdspace.mit.edu/bitstream/handle/1721.1/36896/16...•• The promise of auThe promise of autonomous explorerstonomous explorers •• The challenge of autonomous explorersThe
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
Brian C. Williams
16.410/16.413
September 3rd, 2003
Introduction to Principles of Autonomy and Decision Making
1
Brian C. Williams
16.410/16.413
September 3rd, 2003
Today’s AssignmentToday’s Assignment
•• RRead Chapters 1 and 2 of AIMAead Chapters 1 and 2 of AIMA–– ““ArtificiaArtificial Intelligencl Intelligencee: A Modern Approach”: A Modern Approach”
by Stuart Russell and Peter Norvigby Stuart Russell and Peter Norvig
–– AIMA is available at the CoopAIMA is available at the Coop
1
OutlineOutline
•• The promise of auThe promise of autonomous explorerstonomous explorers•• The challenge of autonomous explorersThe challenge of autonomous explorers•• Agents great and sAgents great and smmallall•• Course objectiCourse objectivve 1 (e 1 (116.410/13):6.410/13):
–– PrinciplePrinciples for Building Agentss for Building Agents
•• CCourse objectiourse objectivve 2 (e 2 (116.413):6.413):– Building auilding ann Agent:Agent:– B
The Mars exploration rover (MER) project.The Mars exploration rover (MER) project.
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov Courtesy of Kanna Rajan.
OutlineOutline
•• The promise of autonomous explorersThe promise of autonomous explorers•• The challenge of autonomous explorersThe challenge of autonomous explorers •• Agents great and smallAgents great and small •• Course objective 1 (16.410/13):Course objective 1 (16.410/13):
–– Principles for Building AgentsPrinciples for Building Agents
•• Course objective 2 (16.413):Course objective 2 (16.413): – Building n Agent:– Building an Agent:
The Mars exploration rover (MER) project.The Mars exploration rover (MER) project.
7
A Capable Robotic Explorer: Cassini • 7 year cruise Faster, Better, Cheaper
•• TThe promise of auhe promise of autonomous explorerstonomous explorers•• TThe challenge of autonomous explorershe challenge of autonomous explorers•• AAgents great and sgents great and smmallall•• CCourse objectiourse objectivve 1 (e 1 (116.410/13):6.410/13):
–– PrinciplePrinciples for Building Agentss for Building Agents
•• CCourse objectiourse objectivve 2 (e 2 (116.413):6.413):–– BBuilding auilding ann Agent:Agent:
The Mars exploration rover (MER) project.The Mars exploration rover (MER) project.
Agents and IntelligenceAgents and Intelligence
Adaoted from J. Malik, U.C. Berkeley
10
Reflex agentsReflex agents
Adaoted from J. Malik, U.C. Berkeley
Goal-oriented agentGoal-oriented agent
Adaoted from J. Malik, U.C. Berkeley
11
Utility-based agentUtility-based agent
Adaoted from J. Malik, U.C. Berkeley
OutlineOutline
•• TThe promise of auhe promise of autonomous explorerstonomous explorers•• TThe challenge of autonomous explorershe challenge of autonomous explorers•• AAgents great and sgents great and smmallall•• CCourse objectiourse objectivve 1 (e 1 (116.410/13):6.410/13):
–– PrinciplePrinciples for Building Agentss for Building Agents
•• CCourse objectiourse objectivve 2 (e 2 (116.413):6.413):–– BBuilding auilding ann Agent:Agent:
The Mars exploration rover (MER) project.The Mars exploration rover (MER) project.
12
Course Objective 1:Course Objective 1: Principles of AgentsPrinciples of Agents
16.410/13: To learn the modeling and16.410/13: To learn the modeling and algorithmic building blocks for creatingalgorithmic building blocks for creating reasoning, learning agents:reasoning, learning agents:
•• TTo formulate reasoning problems.o formulate reasoning problems.•• TTo describe, analyze and demonstrateo describe, analyze and demonstrate
reasoning algorithms.reasoning algorithms.•• TTo model and encodeo model and encode knowledge used byknowledge used by
reasoning algorithms.reasoning algorithms.
Agent ParadigmsAgent Paradigms
• Extensive Reasoning
• Extensive Learning
• Extensive Optimization
• Extensive Reasoning
• Extensive Learning
• Extensive Optimization
13
Extensive Reasoning: Houston, we have a problem ...
•• Planning and Acting in the WPlanning and Acting in the Worldorld•• Constraints and SchedulingConstraints and Scheduling•• Model-based DiagnosisModel-based Diagnosis•• Logic and DeductionLogic and Deduction
Extensive Learning: TD-Gammon [Tesauro, 1995]
Extensive Learning: TD-Gammon [Tesauro, 1995]
Learns to play Backgammon
Situations: • Board configurations (1020)
Actions: • Moves
Rewards: – +100 if win – - 100 if lose – 0 for all other states
• Trained by playing 1.5 million games against self.
Î Currently, roughly equal to best human player.
Learns to play Backgammon
Situations:• Board configurations (1020)
Actions:• Moves
Rewards:– +100 if win– - 100 if lose– 0 for all other states
• Trained by playing 1.5 million games against self.
Î Currently, roughly equal to best human player.
15
Learning methods that supportLearning methods that support the creation of agentsthe creation of agents
•• LLearning through reinforcementearning through reinforcement
ii <= -yma<= -yma xx + Mti4+ Mti4 (for all time i)(for all time i)
6 tik6 tik <= 3<= 3 ((t introduce IP element)t introduce IP element)
•• SimSimilar equation for Collision Avoidance (for all pairs ofilar equation for Collision Avoidance (for all pairs of vehicles)vehicles)
17
OutlineOutline
•• TThe promise of auhe promise of autonomous explorerstonomous explorers•• TThe challenge of autonomous explorershe challenge of autonomous explorers•• AAgents great and sgents great and smmallall•• CCourse objectiourse objectivve 1 (e 1 (116.410/13):6.410/13):
–– PrinciplePrinciples for Building Agentss for Building Agents
•• CCourse objectiourse objectivve 2 (e 2 (116.413):6.413):–– BBuilding auilding ann Agent:Agent:
The Mars exploration rover (MER) project.The Mars exploration rover (MER) project.
Optimization methods thatOptimization methods that support the creation of agentssupport the creation of agents
Mixed Integer Linear Programminging–– Mixed Integer Linear Programm
Solution Methods:Solution Methods:– Dyna– Dynamic Programmingmic Programming– Simplex– Simplex – Branch and Bound– Branch and Bound
18
Course Objective 2:Course Objective 2: Building AgentsBuilding Agents
16.413: To appreciate the challenges of building a16.413: To appreciate the challenges of building a state of the art autonomous explorer:state of the art autonomous explorer:
•• To model and encode knowledge needed to solveodel and encode knowledge needed to solveTo m a state of the art challenge.a state of the art challenge.
•• To work through theTo work through the process of autonomy systemssprocess of autonomy system integration.integration.
•• To assess the promise, frustrations and challengesise, frustrations and challengesTo assess the prom of using (b)leading art technologies.of using (b)leading art technologies.