Online Control of Simulated Humanoids Using Particle Belief Propagation.

Online Control of Simulated Hu-manoids Using Particle Belief Propaga-

tion

Motivation

• Control simulated humanoid• Various movements, environment• Without any pre-computation, motion capture data• At real time

Simulation Model

: State ( pose and velocity )

: Control. ( Desire joint angle )

Character model ( 15 bones, 30 DOF )

Objective

( for balancing )

vel : speed of COMcom : horizontal distance of COM from the feety : y position of COM relative to feetfeet : distance between each footw : angular speed of the pelvisup : difference between the pelvis up vector and global up vectorfwd : head facing directiondamage : 10000 if the character’s head touches the environment

Previous work

Reference Motion

Simula-tion

Fall down

Previous work

Reference Motion

Change reference motion

Previous work

Reference Motion

Simula-tion

Result Control

Simula-tion

Initial Pose

Result Control

Simula-tion

Initial Pose

Pick best sam-ple

Result Control

Simula-tion

Result Control

Simula-tion

Result Control

Simula-tion

Result Control

Simula-tion

N : # of samples ( = 32 )

K : Planning horizon ( = 1.2s, 36 time step )

Simulation Model

: State ( pose and velocity )

: Control. ( Desire joint angle )

Character model ( 15 bones, 30 DOF )

Sampling

Resampling

Backwards local refine-ment

Using previous trajectories as a prior

Probability Model

Probability Model

Probability Model

Control as Markov Random Field

Belief Propagation

Particle Belief Propagation

Particle Belief Propagation

Particle Belief Propagation

Result Control

Simula-tion

Resampling

Local Refinement

Operation Over Multiple Frames

Current Step

Previous Step

Sampling

Operation Over Multiple Frames

Current Step

Previous Step

Total Algorithm