l1 intro 9 rpdspace.mit.edu/bitstream/handle/1721.1/36896/16...•• the promise of authe promise...
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
– 2nd Edition (not 1st Edition!!)– 2nd Edition (not 1st Edition!!)
–– 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.
2
.
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov
A
Image taken from NASA's website: http://www.nasa.gov Courtesy of Kanna Rajan.
Frontier . . .
spacecraft.’’ establish a virtual presence, in space, on planets, in aircraft and ``Our vision in NASA is to open the Space We must
- Daniel S. Goldin, NASA Administrator, May 29, 1996
3
Motive: Follow the water ….
• to find evidence of past life on Mars.
• to study the forces that shaped Mars.
• to develop future life on Mars.
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov Courtesy of Kanna Rajan.
MarsMars
Motive: Follow the water ….
• to find evidence of past life on Mars.
• to study the forces that shaped Mars.
• to develop future life on Mars.
Courtesy U.S. Geological Survey.
4
Inner and Outer Planets Missions
MESSENGER mission to Mercury
MESSENGERmission to Mercury
Venus Sample Return
VenusSample Return
Comet Nucleus Sample Return
Primitive Bodies MissionsPrimitive Bodies Missions
Pluto/Kuiper Express
Europa Orbiter
Europa Lander
Neptune Orbiter
Titan Explorer
Cryobot & Hydrobot
Motive: life underMotive: life under EuropaEuropa??
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov Courtesy of Kanna Rajan.
5
MIT SPHERES Image taken from NASA's website. http://www.nasa.gov. Adapted. Courtesy of Kanna Rajan.
6
Cooperative Exploration
Distributed Planning Group, JPL Model-based Embedded
and Robotic Systems Group, MIT
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
• ~ 150 - 300 ground
operators
•~ 1 billion $ •150 million $
• 7 years to •2 year build build • 0 ground
ops
Four launches in 7 months
Mars Climate Orbiter: 12/11/98 /99
Stardust: 2/7/99 QuickSCAT: 6/19/98
Mars Polar Lander: 1/3
courtesy of JPL
8
Mars Polar Lander
L
Spacecraft require a good physical commonsense…
Launch: 1/3/99
Traditional spacecraft commanding
2; 6;
,
Whats a better paradigm?
GS,SITURN,490UA,BOTH,96-355/03:42:00.000; CMD,7GYON, 490UA412A4A,BOTH, 96-355/03:47:00:000, ON; CMD,7MODE, 490UA412A4B,BOTH, 96-355/03:47:02:000, INT; CMD,6SVPM, 490UA412A6A,BOTH, 96-355/03:48:30:000, CMD,7ALRT, 490UA412A4C,BOTH, 96-355/03:50:32:000, CMD,7SAFE, 490UA412A4D,BOTH, 96-355/03:52:00:000, UNSTOW; CMD,6ASSAN, 490UA412A6B,BOTH, 96-355/03:56:08:000, GV,153,IMM,231,
GV,153; CMD,7VECT, 490UA412A4E,BOTH, 96-355/03:56:10.000, 0,191.5,6.5,
0.0,0.0,0.0, 96-350/ 00:00:00.000,MVR;
SEB,SCTEST, 490UA412A23A,BOTH, 96-355/03:56:12.000 SYS1,NPERR; CMD,7TURN, 490UA412A4F,BOTH, 96-355/03:56:14.000, 1,MVR; MISC,NOTE, 490UA412A99A,, 96-355/04:00:00.000, ,START OF TURN;,
CMD,7STAR, 490UA412A406A4A,BOTH 96-355/04:00:02.000, 7,1701, 278.813999,38.74;
CMD,7STAR, 490UA412A406A4B,BOTH, 96-355/04:00:04.000, 8,350,120.455999, -39.8612;
CMD,7STAR, 490UA412A406A4C,BOTH, 96-355/04:00:06.000, 9,875,114.162, 5.341;
CMD,7STAR, 490UA412A406A4D,BOTH, 96-355/04:00:08.000, 10,159,27.239, 89.028999;
CMD,7STAR, 490UA412A406A4E,BOTH, 96-355/04:00:10.000, 11,0,0.0,0.0; CMD,7STAR, 490UA412A406A4F,BOTH, 96-355/04:00:12.000, 21,0,0.0,0.0;
9
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.
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 ...
courtesy of NASA
• Quintuple fault occurs (three shorts, tank-line and pressure jacket burst, panel flies off).
• Mattingly works in ground simulator to identify new sequence handling severe power limitations.
• Mattingly identifies novel reconfiguration, exploiting LEM batteries for power.
• Swaggert & Lovell work on Apollo 13 emergency rig lithium hydroxide unit.
Image taken from NASA's website: http://www.nasa.gov. Courtesy of Kanna Rajan.
Example of a Model-based Agent:
• Goal-directed
• First time correct
• projective • reactive
• Commonsense models
• Heavily deductive
Scripts
component models
GoalsGoals
DiagnosisDiagnosis& Repair& Repair
MissionMissionManagerManager ExecutiveExecutive
Planner/Planner/SchedulerScheduler
Remote AgentRemote Agent
Mission-level actions & resources
14
Reasoning methods thatReasoning methods that support the creation of agentssupport the creation of agents
•• Rule-bRule-based systemsased systems–– ForwaForward chainingrd chaining–– GoalGoal-directeddirected-
•• Search and roadmSearch and roadmap path planningap path planning–– Blind SearchBlind Search –– InformInformed Searched Search–– AdversaAdversarial (Game) Searchrial (Game) Search
•• 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
•• LLearearnining decision treesng decision trees
•• NNeural net learningeural net learning
Extensive OptimizationExtensive Optimization
16
Cooperative Path Planning: MILP Encoding: Fuel Equation
Cooperative Path Planning:MILP Encoding: Fuel Equation
min = JT = min 6 q’wi + 6 r’vi + p’wNmin = JT = min 6 q’wi + 6 r’vi + p’wNwi, vi wi, vi i=1
N-1
i=1
N-1
slack control vector weighting vectors
slack state vector
past-horizon terminal cost term
total fuel calculated over all time instants i
Cooperative Path Planning:Cooperative Path Planning:MILP Encoding: ConstraintsMILP Encoding: Constraints
• s• sijij <= w<= wij, etc.ij, etc. State Space ConstraintsState Space Constraints
•• ssii+1 = As+1 = Asii + Bu+ Buii State Evolution EquationState Evolution Equation
-x-x• x• xii <= xmi<= xmi n + Mti1+ Mti1n
ii <= -x<= -x + Mti2+ Mti2mama xx
-y-yyyii <= ymi<= ymi n + Mti3+ Mti3n Obstacle AvoidanceObstacle Avoidance
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
Modeling Frameworks:Modeling Frameworks:– Markov Decision Processes– Markov Decision Processes
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.
Mars Exploration Rovers – Jan. 2004Mars Exploration Rovers – Jan. 2004
Mission Objectives:
• Learn about ancient water and climate on Mars.
• For each rover, analyze a total of 6-12 targets – Targets = natural rocks, abraded rocks, and soil
• Drive 200-1000 meters per rover
• Take 1-3 panoramas both with Pancam and mini-TES
• Take 5-15 daytime and 1-3 nightime sky observations with mini-TES
Mission Objectives:
• Learn about ancient water and climate on Mars.
• For each rover, analyze a total of 6-12 targets– Targets = natural rocks, abraded rocks, and soil
• Drive 200-1000 meters per rover
• Take 1-3 panoramas both with Pancam and mini-TES
• Take 5-15 daytime and 1-3 nightime sky observations with mini-TES
Mini-TES Pancam
Navcam
Rock Abrasion Tool Microscopic Imager
Mossbauer spectrometer APXS
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov. Courtesy of Kanna Rajan.
19
Surface Operations Scenario
Day 4 Day 1
Day 2
Board Navigation Changes, as needed
Day 3
Mars Exploration Rover
Target
During the Day Science Activities
Long-Distance Traverse (<20-50 meters)
Initial Position; Followed by “Close Approach”
During the Day Autonomous On-
Day 2 Traverse Estimated Error Circle
Science Prep (if Required)
Day 2 Traverse Estimated Error Circle
Courtesy of Kanna Rajan, NASA Ames. Used with permission.
Activity Name Durati
on 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 6 7 8 9
DTE 4.50 0.75
DTE period DFE
Night Time Rover Operations 16.97 Night Time Rover OperationsSleep Wakeup
Pre-Comm Session Sequence Plan Review
Current Sol Sequence Plan Review 1.50 1.50
Current Sol Sequence Plan Review
Prior Sol Sequence Plan Review 2.00 Prior Sol Sequence Plan Review
Real-TIme Monitoring 4.50 0.75
Real-TIme Monitoring Real-TIme Monitoring
2.75 Downlink Product Generation
Tactical Science Assessment/Observation Planning
5.00 Tactical Science Assessment/Observation Planning
Science DL Assessment Meeting 1.00 Science DL Assessment Meeting
Payload DL/UL Handoffs 0.50 Payload DL/UL Handoffs
Tactical End-of-Sol Engr. Assessment & Planning
5.50 Tactical End-of-Sol Engr. Assessment & Planning
DL/UL Handover Meeting 0.50 DL/UL Handover Meeting
Skeleton Activity Plan Update 2.50 Skeleton Activity Plan Update
SOWG Meeting 2.00 SOWG Meeting
Uplink Kickoff Meeting 0.25 Uplink Kickoff Meeting
Activity Plan Integration & Validation 1.75 Activity Plan Integration & Validation
Activity Plan Approval Meeting 0.50 Activity Plan Approval Meeting
Build & Validate Sequences 2.25 Build & Validate Sequences
UL1/UL2 Handover 1.00 UL1/UL2 Handover
Complete/Rework Sequences 2.50 Complete/Rework Sequences
Margin 1 0.75 Margin 1
Command & Radiation Approval 0.50 Command & Radiation Ap
Margin 2 1.25 Margin 2
Radiation 0.50 Radiation
MCT Team 7.00 4.00
One day in the life of a Mars rover
Science Planning Sequence Build/Validation Uplink
Downlink Product Generation...
Courtesy: Jim Erickson
Downlink Assessment
Courtesy of Kanna Rajan, NASA Ames. Used with permission.
20
EUROPAAutomated
Planning System
EUROPA Automated
Planning System
Science
Navigation
Engineering
Resource Constraints
DSN/Telcom
Flight Rules
Science Team
Sequence Build
MAPGEN: Automated Science Planning for MER
Planning Lead: Kanna Rajan (ARC)
Courtesy of Kanna Rajan, NASA Ames. Used with permission.
Course ChallengeCourse Challenge
• What would it be like to operate MER if it was fully autonomous?
Course project:
• Demonstrate an autonomous MER mission in simulation, and in the MIT rover testbed.
• What would it be like to operate MER if it was fully autonomous?
Course project:
• Demonstrate an autonomous MER mission in simulation, and in the MIT rover testbed.
21
Next Challenge: Mars Smart Lander (2009)
Next Challenge: Mars Smart Lander (2009)
Mission Duration: 1000 days Total Traverse: 3000-69000 meters Meters/Day: 230-450 Science Mission: 7 instruments, sub-surface science package (drill, radar), in-situ sample “lab”
Technology Demonstration: (2005).
Courtesy NASA/JPL-Caltech. http://www.jpl.nasa.gov. Courtesy of Kanna Rajan.
22