Transcript High Level Motion Control Slides in PPT.

CS274: Computer Animation and Simulation
Lecture V
Higher Level Motion Control
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Higher Level Motion Control
We often wish to specify higher level goals
rather than joint angles and translations
(Semi-)autonomous creatures reduce animator
load and improve interactive applications
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Control Algorithms
Control algorithms translate high level objectives
into motor controls and joint angles
Useful for motions like walking created by
several coordinated muscle actions
We maintain balance, speed, etc. by continually
making small adjustments based on the situation
Try to mimic what works naturally!!!
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Control Algorithms
Simplified control loop
User
Control
Simulation
Frame
Use feedback to maintain:
 balance
 velocity (speed and direction)
 etc.
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Control Laws
How do we determine control laws?
 By hand
 Biomechanics data
 Optimization
Motions like walking, running, etc. can be broken
into smaller sections that are easier to analyze
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
State Machines
Separate the motion into several simple states
Simple states allow us to generate laws by hand
State transitions are triggered by events
Example: fall forward until foot hits the ground
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Events
Often simple binary sensors
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Running State Machine
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Flight Stage
Active Leg
 swing leg forward
 straighten knee
Passive Leg
 mirror active hip angle
 bend knee
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Heel Contact Stage
Active Leg
 pitch/balance control with hip
 extend ankle
 knee acts like a spring (thrust in next stage)
Passive Leg
 mirror active hip angle
 bend knee
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Secondary Control Laws
To add realism
 waist keeps body upright
 neck facing desired direction
 shoulder mirrors hip angle
 elbow angle is a function of
shoulder angle
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Low Level Control
How do we get the knee to “hold” or “extend”?
Again, mimic the muscle actions at a joint
Similar to a damped spring
These muscle motors are known as actuators
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Actuators
Actuators make the joint move to a desired pose
Angular spring
τ  k ( d   )  kv (d  )
Linear spring
f  k (Ld  L )  kv ( Ld  L )
Also known as proportional derivative controllers
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Walk Cycle
No flight phase in a walk cycle
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Walk Cycle
Walk cycle timeline
Double
support
Left support
Left stance
Right swing
CS274 Spring 01 Lecture 5
Double
support
Right support
Left swing
Right stance
Copyright © Mark Meyer
Results
Olympic Running
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Results
Vaulting, Cycling and Running
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Results
Combining controllers and retargetting
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Self-taught creatures?
 Sensor-Actuator networks:
 Using the same basic tools
 Try to find the right coefficients to maximize
speed, or efficiency, or any energy.
 Your humanoid learns to run on his own!
 Slow, though…
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Self Taught Creatures
Creatures that can learn to move on their own
Consists of sensor-actuator networks
 sensors tell the character about the environment
 actuators allow the character to flex its muscles
As the character moves, it remembers muscle
movements that create “good” motion
 maximize speed, accuracy, efficiency, etc.
t2
E   1 Eactuators   2 Emotion dt
t1
measures energy used
CS274 Spring 01 Lecture 5
measures motion quality
Copyright © Mark Meyer
Spacetime Constraints
Solve for the forces required to reach constraints
Animator specifies:
What the character has to do (constraints)
 initial, intermediate, final positions, velocities, etc.
How the motion should be performed (metric)
 jump this high, this much force at impact
The character’s physical structure
 mass, joints, etc.
What the physical resources are
 constraints on the muscles
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Spacetime Constraints
Given the constraints
 Animator specified constraints
 Muscle and joint constraints
 Physical laws
And the metric to optimize
 How to perform the action
Solve the constrained optimization for the force
curves over the entire time of the motion
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Behavioral/Procedural Animation
Specify behaviors that the actor should follow
Complexity emerges from multiple interactions
Separation
Avoid crowding
CS274 Spring 01 Lecture 5
Alignment
Match velocity
Cohesion
Match centroid
Copyright © Mark Meyer
Behaviors
Seek/Flee
Adjust for a radial velocity
Arrival
Seek and decelerate near target
Other variants: pursuit, evasion, offset pursuit
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Behaviors
Obstacle Avoidance
Keep the cylinder in front clear
CS274 Spring 01 Lecture 5
Wander
Smooth random motion
Copyright © Mark Meyer
Behaviors
Path Following
Follow a small tube
CS274 Spring 01 Lecture 5
Wall Following
Path follow + Obstacle Avoidance
Copyright © Mark Meyer
Behavioral/Procedural Animation
CS274 Spring 01 Lecture 5
Useful for crowd animations
Copyright © Mark Meyer
Cognitive Modeling
Use AI to allow for planning and learning
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Putting It All Together
Behavior + Learning + Motor Control
Schooling
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Putting It All Together
Behavior + Learning + Motor Control
Preying
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer
Putting It All Together
Behavior + Learning + Motor Control
Cousto World
CS274 Spring 01 Lecture 5
Copyright © Mark Meyer