Big Picture Assignment four What is your big

Big Picture • Assignment four • What is your big picture goal? Breaking It Down • Implementing your goal – Behaviors • Model Plan Act • Emergent Behavior • Finite State Machines

Assignment Four • Write-Up (under a page) • Rough Sketches • Rough Schedule

Assignment Four – Write-Up • What are the main design features? • What is special about your robot? • Strategy • Find the ball/goal • Go to the ball/goal • Pick up/release the ball • Physical Design • Hardware • Motors • Sensors • Keep track of sensor points • Software Design • Code organization • Vision Processing

Assignment Four - Sketches • Not elaborate • Labeled

Assignment Four - Schedules • Not elaborate • Fill in the details (what behaviors, etc. ) – not generic

Big Picture • Assignment four • What is your big picture goal? Breaking It Down • Implementing your goal – Behaviors • Model Plan Act • Emergent Behavior • Finite State Machines

What is your eventual plan for your robot? Collect balls Score points Wander Find goals …

Big Picture • Assignment four • What is your big picture goal? Breaking It Down • Implementing your goal – Behaviors • Model Plan Act • Emergent Behavior • Finite State Machines

How can you break down your goal into behaviors? ? Behavior: a set or series of acts regarded as a unified whole

What are examples of a behavior: • Move towards ball • Turn right • Move away from wall

How can you coordinate behaviors? Model Emergent Finite State Plan Act Behavior Machine Choose between behaviors based on inputs Switch between the behaviors based on inputs and past behavior Execute a sequence of behaviors designed to deal with a model of the situation

Managing Behaviors – Model Plan Act 1. Model 2. Plan 3. Act Create world model Use the model to sequence behaviors to accomplish goal Execute plan (sensors) (actuators)

Managing Behaviors – Model Plan Act Model Plan Sensors Actuators Environment

Managing Behaviors – Model Plan Act Model: traversability line of sight graph

Managing Behaviors – Model Plan Act Plan/Act

Managing Behaviors – Model Plan Act New model

Managing Behaviors – Model Plan Act Good: Optimize behavior Bad: All behaviors must be complete before testing Takes time to make a model What if model is no good? (noisy sensors, dynamic environment) Video 1

How can you coordinate behaviors? Model Emergent Finite State Plan Act Behavior Machine Choose between behaviors based on inputs Switch between the behaviors based on inputs and past behavior Execute a sequence of behaviors designed to deal with a model of the situation

Managing Behaviors – Emergent behavior Arbitration between simple behaviors can lead to sophisticated results. Camera IR Go to ball Avoid Wander Arbiter Actuators

Managing Behaviors – Emergent behavior Arbitration between simple behaviors can lead to sophisticated results. • Fixed priority • Round-Robin • Vote Camera IR • Merge Go to ball • Random Avoid Wander Arbiter Actuators

Managing Behaviors – Emergent behavior Arbitration between simple behaviors can lead to sophisticated results. Camera IR Go to ball Avoid Wander Arbiter Actuators

Managing Behaviors – Emergent behavior Arbitration between simple behaviors can lead to sophisticated results. Ball Switch Camera IR Hold ball Arbiter Go to goal Gate Motor Go to ball Avoid Wander Arbiter Actuators

Managing Behaviors – Emergent behavior What behaviors are there? Video 2 (from 0: 37)

Managing Behaviors – Emergent behavior Freak out Camera Charge goal Chase ball IR Go to open areas Arbiter Actuators

Managing Behaviors – Emergent behavior No behavior for: • Resume after getting ball • Retreat from goal • Wander Freak out Camera Charge goal Chase ball IR Go to open areas Arbiter Actuators

Managing Behaviors – Emergent Behavior Why use this? • Development is natural • Modularity • Fine with dynamic environments Why not use this? • Hard to predict Video 2 • Debugging hard (from 1: 30)

How can you coordinate behaviors? Model Emergent Finite State Plan Act Behavior Machine Choose between behaviors based on inputs Switch between the behaviors based on inputs and past behavior Execute a sequence of behaviors designed to deal with a model of the situation

Finite State Machines Switch between behaviors based on current state and inputs. Some behaviors rely on some planning and state. Some transitions determined statically, some dynamically.

Finite State Machines Switch between behaviors based on current state and inputs. Some behaviors rely on some planning and state. Some transitions determined statically, some dynamically. Detect ball Chase Ball Detect ball See no ball Turn 360 30 seconds Wander Random

Finite State Machines Can get very complicated … can have FSMs within states.

Finite State Machines Advantages • Logical • Modular - Easy to improve, test, add • Predictable

Finite State Machines Advantages • Logical • Modular - Easy to improve, test, add • Predictable Disadvantages • Not quite as natural as emergent • Not optimum solution like MPA

Finite State Machines Advantages • Logical • Modular - Easy to improve, test, add • Predictable Video 3 Disadvantages • Not quite as natural as emergent • Not optimum solution like MPA

How can you coordinate behaviors? Model Emergent Finite State Plan Act Behavior Machine Choose between behaviors based on inputs Switch between the behaviors based on inputs and past behavior Execute a sequence of behaviors designed to deal with a model of the situation