Behavior Trees vs FSMs An Introduction to Behavior
Description: Behavior Trees vs FSMs An Introduction to Behavior Trees in AI (part 4) by Petter Ögren Similarities FSM - BT 2022-11-18 2 Both answer the question: What to do next? Grasp Drop Walk Open Close Search Clean Speak Run Recharge Lock Unlock
Related Topics
Download Presentation
"Behavior Trees vs FSMs An Introduction to Behavior" is the property of its rightful owner. Permission is granted to download and print the materials on this website for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Presentation Transcript
slide1. Behavior Trees vs FSMs An Introduction to Behavior Trees in AI (part 4)
by Petter Ögren<br>
slide2. Similarities FSM - BT 2022-11-18 2 Both answer the question: What to do next? Grasp Drop Walk Open Close Search Clean Speak Run Recharge Lock Unlock Pull Push Kick Speak Record Throw<br>
slide3. Pros and Cons 2022-11-18 3 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy problem
State feedback is optimal
Next Action should only depend on current World State, not on current Action (FSM state) Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Good for smaller problems! Good for larger problems!<br>
slide4. FSMs 2022-11-18 4 Grasp Drop Walk Open Close Search Clean Speak Recharge Lock Unlock Pull Push Kick Speak Record Finite State Machine Each action needs to know “What to do next”…<br>
slide5. Pros and Cons 2022-11-18 5 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy
State feedback is optimal
Next Action should only depend on current System State, not on current Action<br>
slide6. GOTO analogy 2022-11-18 6 Problem 1: (50 years ago)
“The GOTO statement should be abolished from all higher level programming languages”
Paper by E. Dijkstra, 1968
Reason: GOTO harms Modularity
Solution: GOTO not used anymore
Problem 2:
Finite State Machines are full of GOTO statements (transitions)<br>
slide7. State Feedback is Optimal 2022-11-18 7 From Reinforcement Learning
The optimal policy is a function of the world state s
It does not depend on the previous action (the internal FSM-state)
The optimal transition structure is identical for all FSM-states
A degenerate FSM… s<br>
slide8. Pros and Cons 2022-11-18 8 Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Behavior Tree Each action needs to know “Did I Succeed or Fail?” Grasp (parent) (tick) Success,
Failure,
or Running<br>
slide9. Pros and Cons 2022-11-18 9 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy
State feedback is optimal
Next Action should only depend on current World State, not on current Action (FSM state) Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Good for smaller problems! Good for larger problems! Designed by pch.vector / Freepik Bye!<br>
by Petter Ögren<br>
slide2. Similarities FSM - BT 2022-11-18 2 Both answer the question: What to do next? Grasp Drop Walk Open Close Search Clean Speak Run Recharge Lock Unlock Pull Push Kick Speak Record Throw<br>
slide3. Pros and Cons 2022-11-18 3 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy problem
State feedback is optimal
Next Action should only depend on current World State, not on current Action (FSM state) Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Good for smaller problems! Good for larger problems!<br>
slide4. FSMs 2022-11-18 4 Grasp Drop Walk Open Close Search Clean Speak Recharge Lock Unlock Pull Push Kick Speak Record Finite State Machine Each action needs to know “What to do next”…<br>
slide5. Pros and Cons 2022-11-18 5 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy
State feedback is optimal
Next Action should only depend on current System State, not on current Action<br>
slide6. GOTO analogy 2022-11-18 6 Problem 1: (50 years ago)
“The GOTO statement should be abolished from all higher level programming languages”
Paper by E. Dijkstra, 1968
Reason: GOTO harms Modularity
Solution: GOTO not used anymore
Problem 2:
Finite State Machines are full of GOTO statements (transitions)<br>
slide7. State Feedback is Optimal 2022-11-18 7 From Reinforcement Learning
The optimal policy is a function of the world state s
It does not depend on the previous action (the internal FSM-state)
The optimal transition structure is identical for all FSM-states
A degenerate FSM… s<br>
slide8. Pros and Cons 2022-11-18 8 Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Behavior Tree Each action needs to know “Did I Succeed or Fail?” Grasp (parent) (tick) Success,
Failure,
or Running<br>
slide9. Pros and Cons 2022-11-18 9 Finite State Machine (FSM)
Pros
Easy to implement
Easy to understand
Cons
Does not scale well (n^2 transitions)
GOTO analogy
State feedback is optimal
Next Action should only depend on current World State, not on current Action (FSM state) Behavior Tree (BT)
Pros
Modular (weak dependence between subtrees)
Function call analogy
Generalizes
Decision Trees
Subsumption architecture
Teleo-reactive approach
Cons
Somewhat harder to implement
Somewhat harder to understand Good for smaller problems! Good for larger problems! Designed by pch.vector / Freepik Bye!<br>