CS325 Artificial Intelligence Ch. 11, Advanced Planning Cengiz - PowerPoint PPT Presentation

CS325 Artificial Intelligence Ch. 11, Advanced Planning Cengiz Günay, Emory Univ. Spring 2013 Günay Ch. 11, Advanced Planning Spring 2013 1 / 12

Advanced Planning Concepts This lecture: Time: Scheduling Günay Ch. 11, Advanced Planning Spring 2013 2 / 12

Advanced Planning Concepts This lecture: Time: Scheduling Resources: Consumables Günay Ch. 11, Advanced Planning Spring 2013 2 / 12

Advanced Planning Concepts This lecture: Time: Scheduling Resources: Consumables Active perception: Look and feel? Günay Ch. 11, Advanced Planning Spring 2013 2 / 12

Advanced Planning Concepts This lecture: Time: Scheduling Resources: Consumables Active perception: Look and feel? Hierarchical plans: Abstracting Günay Ch. 11, Advanced Planning Spring 2013 2 / 12

Entry/Exit Surveys Exit survey: Game Theory Why don’t we take the mixed strategy if there is a dominant strategy? What advantage is gained by a player by looking irrational? Entry survey: Advanced Planning (0.25 points of final grade) Do you think a classical planning planning approach can be used for solving scheduling problems? What would be the advantage of making hierarchical plans? Günay Ch. 11, Advanced Planning Spring 2013 3 / 12

I’m Late Again! Fixed times for day start and class Durations Wake up:10 minutes Eat: 30 minutes Start (8am) Wake up Class (10am) Eat Günay Ch. 11, Advanced Planning Spring 2013 4 / 12

I’m Late Again! Fixed times for day start and class Durations Wake up:10 minutes Eat: 30 minutes Start (8am) Wake up Class (10am) Eat Earliest and latest start times of each event? Günay Ch. 11, Advanced Planning Spring 2013 4 / 12

I’m Late Again! Fixed times for day start and class Durations Wake up:10 minutes Eat: 30 minutes Start (8am) Wake up Class (10am) Eat Earliest and latest start times of each event? Earliest(Wake up)=8am Latest(Wake up)=? Earliest(Eat)=? Latest(Eat)=10:00 − 00:30 = 9:30 am Günay Ch. 11, Advanced Planning Spring 2013 4 / 12

Multiple Paths to Finish: Car with 2 Engines Critical path? AddWheels1 AddEngine1 Inspect1 AddEngine2 Inspect2 AddWheels2 0 10 20 30 40 50 60 70 80 90 Günay Ch. 11, Advanced Planning Spring 2013 5 / 12

Multiple Paths to Finish: Car with 2 Engines AddWheels1 AddEngine1 Inspect1 AddEngine2 Inspect2 AddWheels2 0 10 20 30 40 50 60 70 80 90 [ 0 , 1 5] [ 3 0, 4 5] [60,75] AddEngine1 AddWheels1 Inspect1 30 30 10 [0,0] [85,85] Star t Finish [0,0] [60,60] [75,75] AddEngine2 AddWheels2 Inspect2 60 15 10 Günay Ch. 11, Advanced Planning Spring 2013 5 / 12

Multiple Paths to Finish: Car with 2 Engines [ 0 , 1 5] [ 3 0, 4 5] [60,75] AddEngine1 AddWheels1 Inspect1 30 30 10 [0,0] [85,85] Star t Finish [0,0] [60,60] [75,75] AddEngine2 AddWheels2 Inspect2 60 15 10 Earliest(Start)=0 Earliest( B )=max A → B Earliest ( A ) + Duration ( A ) Latest( A )=max A ← B Latest ( B ) − Duration ( A ) Latest(Finish)=Earliest(Finish) Günay Ch. 11, Advanced Planning Spring 2013 5 / 12

Resources: Can We Use Action Schemas? Günay Ch. 11, Advanced Planning Spring 2013 6 / 12

Resources: Can We Use Action Schemas? Will it reach the goal? Günay Ch. 11, Advanced Planning Spring 2013 6 / 12

Resources: Can We Use Action Schemas? Will it reach the goal? No , one nut is missing. Günay Ch. 11, Advanced Planning Spring 2013 6 / 12

Resources: Can We Use Action Schemas? Will it reach the goal? No , one nut is missing. Depth first tree search: how many paths to eval? Small: 1, 4, 5 ? Medium: 4 + 5 or 4 × 5 ? Large: 4 ! , 5 ! , or 4 ! × 5 ! ? Günay Ch. 11, Advanced Planning Spring 2013 6 / 12

Resources: Can We Use Action Schemas? Will it reach the goal? No , one nut is missing. Depth first tree search: how many paths to eval? Small: 1, 4, 5 ? Medium: 4 + 5 or 4 × 5 ? Large: 4 ! , 5 ! , or 4 ! × 5 ! . Really inefficient! Günay Ch. 11, Advanced Planning Spring 2013 6 / 12

Modify Action Schemas to Optimize Resource Problems No need to try combinations of same resources. Define: Resources: Specify quantity. Use: Specify requirement. Consume: Removes resource. Günay Ch. 11, Advanced Planning Spring 2013 7 / 12

Modify Action Schemas to Optimize Resource Problems No need to try combinations of same resources. Define: Resources: Specify quantity. Use: Specify requirement. Consume: Removes resource. Günay Ch. 11, Advanced Planning Spring 2013 7 / 12

Modify Action Schemas to Optimize Resource Problems No need to try combinations of same resources. Define: Resources: Specify quantity. Use: Specify requirement. Consume: Removes resource. No exponential explotion anymore! Günay Ch. 11, Advanced Planning Spring 2013 7 / 12

Hierarchical Planning: Abstraction Remember Stanley: High-level goal: Reach target at GPS coordinates Drive on road Low-level actions: Adjust steering wheel Press/release gas/break pedals Günay Ch. 11, Advanced Planning Spring 2013 8 / 12

Hierarchical Planning: Abstraction Remember Stanley: High-level goal: How to connect 1 high-level (abstract) planning with Reach target at GPS coordinates 2 low-level planning? Drive on road Low-level actions: Adjust steering wheel Press/release gas/break pedals Günay Ch. 11, Advanced Planning Spring 2013 8 / 12

Hierarchical Planning: Abstraction Remember Stanley: High-level goal: How to connect 1 high-level (abstract) planning with Reach target at GPS coordinates 2 low-level planning? Drive on road Solution: Refinement Low-level actions: Adjust steering wheel Press/release gas/break pedals Günay Ch. 11, Advanced Planning Spring 2013 8 / 12

Refining Abstractions Multiple ways to refine abstractions: Günay Ch. 11, Advanced Planning Spring 2013 9 / 12

Hierarchical Planning: Reachable States Reachable? Günay Ch. 11, Advanced Planning Spring 2013 10 / 12

Hierarchical Planning: Reachable States Reachable? No. Günay Ch. 11, Advanced Planning Spring 2013 10 / 12

Hierarchical Planning: Reachable States Reachable? Found solution: No. Günay Ch. 11, Advanced Planning Spring 2013 10 / 12

Hierarchical Planning: Reachable States Reachable? Found solution: No. Now backtrack from solution. Günay Ch. 11, Advanced Planning Spring 2013 10 / 12

Reachable States Question Estimates of refinement: Underestimate: We reach it for sure. Overestimate: Possibly reachable. Below examples: Reachable? Yes, No, Maybe? Günay Ch. 11, Advanced Planning Spring 2013 11 / 12

Reachable States Question Estimates of refinement: Underestimate: We reach it for sure. Overestimate: Possibly reachable. Below examples: Reachable? Yes, No, Maybe? No Yes. Günay Ch. 11, Advanced Planning Spring 2013 11 / 12

Reachable States Question Estimates of refinement: Underestimate: We reach it for sure. Overestimate: Possibly reachable. Below examples: Reachable? Yes, No, Maybe? No Yes. Günay Ch. 11, Advanced Planning Spring 2013 11 / 12

Reachable States Question Estimates of refinement: Underestimate: We reach it for sure. Overestimate: Possibly reachable. Below examples: Reachable? Yes, No, Maybe? No Yes. Maybe. Günay Ch. 11, Advanced Planning Spring 2013 11 / 12

Extending Planning with Observations Sometimes the agent needs to look first. Günay Ch. 11, Advanced Planning Spring 2013 12 / 12

Extending Planning with Observations Sometimes the agent needs to look first. Günay Ch. 11, Advanced Planning Spring 2013 12 / 12