Steering Behavior

1
Steering Behavior

• CIS 488/588
• Bruce R. Maxim
• UM-Dearborn

2
Alife

• Focuses on behavior of individual creatures that
  combine into complex pattern
• The interaction of simple patterns can create
  incredibly sophisticated simulations
• Alife can be used to simulate crowds in complex
  environments using streering behaviors
• This was the basis of Conways game of life

3
Assumptions

• Steering behaviors assume the existence of a
  lower-level of the engine to handle locomotion
• The locomotion system processes each characters
  position and velocity
• When player stops pressing a key movement stops
  (heavy duty friction)
• In many games a key press is likely to control
  velocity and not acceleration
• In this chapter velocity persists and the AI
  applies an acceleration to effect steering

4
Seeking and Fleeing

• Seeking
• Steering behavior moving creatures toward target
• Fleeing
• Steering behavior moving creatures away from
  target
• Pursuit
• Seeking a moving target
• Evasion
• Avoiding a moving target

5
Enhancements - 1

• Wandering
• Make random behavior appear a little more
  purposeful
• Patterns should be slightly unpredictable, but
  not arbitrarily random
• Could accumulate steering values and filter them
  using the sin function (gives both and
  values)
• Projecting Targets
• Look ahead randomly for point where target is
  predicted to be and move there

6
Seeking Implementation

• compute velocity vector toward target
• desired_velocity truncate(position_target,max_sp
  eed)
• compute steering force
• steering_force desired_velocity -
  velocity
• Arrival behavior can be simulated by slowing down
  the velocity to something less than max_speed
  within some distance from target

7
Fleeing Implementation

• // compute velocity vector toward target
• desired_velocity truncate(position_target,max_sp
  eed)
• // compute steering force
• steering_force - desired_velocity -
  velocity

8
Modeling Flocks

• There are additional steering components needed
  (e.g. alignment and cohesion) when modeling group
  behaviors
• These require position and orientation
  information for neighbors rather than surrounding
  obstacles

9
Generalized Obstacle Avoidance

• function avoid_obstacles
• project future position based on velocity
• if collision predicted
• project empty position away from collision
• compute turn to get to empty position
• if obstacle is within critical distance
• determine braking force // slow or stop
• apply steering and braking forces

10
Comments

• Application of steering and braking forces is
  easy since our interface outputs turn and move
  values
• The challenge is in acquiring the inputs used to
  determine future collisions
• This algorithm requires environment knowledge,
  intersection tests, collision normal forces, and
  location of nearby empty spaces
• This makes it unsuitable for implementation

11
Updated Obstacle Avoidance - 1

• function avoid_obstacles2
• check sensors for free space front, left, right
• if front collision predicted
• find furthest obstacle on right or left
• determine best side to turn toward
• compute turn to seek that free space
• if front obstacle is within critical distance
• determine braking force // slow or stop

12
Updated Obstacle Avoidance - 2

• if obstacle on left
• adjust steering to step right
• if obstacle on right
• adjust steering to step left
• apply steering and braking forces
• Implementing this involves simple translation to
  C and finding suitable parameters during
  experimentation phase

13
Enhancements - 2

• Forced Exploration
• Keep track of previous positions a flee from them
• Done using single vector (provenance)pointing
  toward last position
• Coefficients a b 1
• provenance a previous
• b provenance

14
Advantages - 1

• Simplicity
• Hard to find a simpler architecture
• Reliability
• Most situations can be identified requirements
• Predictability
• No ambiguity, each rule written explicitly
• Efficiency
• Very low computational overhead

15
Disadvantages - 1

• Local traps
• Can still get stuck in corners, if it begins to
  turn one way and then decides the another way
  might be better
• Testing
• This approach requires extensive testing because
  there are so many parameters to play with

16
Disadvantages - 2

• Realism
• Robotic movement is not as smooth as it might be
  (local decisions are not integrated into a plan)
• Scalability
• Additional behaviors are added by adding
  additional lines of code and recompiling

17
Marvin

• Uses obstacle sensors and steering behaviors to
  prevent collisions in a reactive fashion
• Uses chapter enhancements to improve its
  wandering behavior