Interactive Motion Correction and Object Manipulation

1
Interactive Motion Correction and Object
Manipulation
ACM SIGGRAPH Symposium on Interactive 3D
Graphics and Games 2007

• Ari Shapiro , UCLA
• Marcelo Kallmann, University of California,
  Merced
• Petros Faloutsos , UCLA

2
Online

• Introduction
• Related Work
• Problem Formulation
• Synchronized Motion Planner
• Inverse Kinematics
• Result
• Conclusion

3
Introduction

• Interactive motion editing approach
• motion correction (to remove collisions)
• synthesis of realistic object manipulation
  sequences on top of locomotion.

4
Introduction

• Motion Correction
• Set times tinit and tgoal such that interval
  spans the problematic period of the motion.
• Solved by planning a new path between tinit and
  tgoal .
• If the planner is successful, the result will be
  a collision-free motion that is used to replace
  the original motion.

5
Introduction

• Interactive Object Manipulation
• Specify hand targets on the objects to be
  grasped.
• A hand target to be reached at a given time tb.
• Determine times ta and tc such that ta lt tb lt tc.
• First path is planned between ta and tb.
• Second path is planned between tb and tc.
• Joint configuration is determined by employing
  the Inverse Kinematics.

6
Motivation

• The motion capture data must be modified to
  accommodate the virtual environment.
• Designing in advance all required motions
  involves tedious and time-consuming design work.
• Introduce a new motion editing approach that
  combines recorded motions with motion planning.

7
Related Work

• Full-body locomotion planning
• A 2-Stages Locomotion Planner for Digital
  Actors. Pettre et al., SCA 2003
• Plan the movement of the character
• Correct the upper body for collisions.
• Behavior planning for character animation.
  Lau and Kuffner, SIGGRAPH 2005
• Uses the time dimension

8
Related Work

• Reach and arm planning
• RRT-Connect An Efficient Approach to
  Single-Query Path Planning. Lavalle et al.,
  ICRA 2000
• IEEE Intl Conf. on Robotics and Automation
• Rapidly-Exploring Random Tree
• Bidirectional version

9
Problem Formulation

• CF be the space of all full Configurations of the
  character.
• DOFs controlled by the planner.
• DOFs controlled by an external motion.

10
Problem Formulation

• An planner controller affecting the DOFs in CP is
  defined as a time-varying function mp(t).

11
Synchronized Motion Planner

• RRT Extend

12
Synchronized Motion Planner

• RRT-Connect

13
Synchronized Motion Planner
14
Synchronized Motion Planner

• Searching for the closest configurations
• be the position of the joint affected by
• rotation
• wt and wa are the desired weights.

15
Synchronized Motion Planner

• Node expansion
• Node interpolation
• the configuration in Tinit has to have its time
  component small than the configuration in Tgoal
• no collision.

16
Configuration Sampling

• Importance in determining the quality of a
  solution and how fast it is found.
• Joint Parameterization
• Joint Limits
• Collision Detection
• Search Heuristics

17
Configuration Sampling

• Joint Parameterization
• Shoulder (3 DOFs)
• swing-and-twist decomposition
• Elbow (2 DOFs)
• flexion and twist rotations
• defined with two Euler angles
• Wrist (2 DOFs)
• swing rotation

18
Configuration Sampling

• Joint Limits
• the swing limits based on spherical ellipses.
• the twist and flexion rotations are limited by
  min and Max angles.

19
Configuration Sampling

• Collision Detection
• VCollide package
• Oriented bounding boxes tree.
• Employed for querying if body parts
  self-intersect or intersect with the environment.

20
Configuration Sampling

• Search Heuristics
• Uniformly sampling valid postures has the effect
  of biasing the search toward the free spaces.
• Biasing method
• Perform a bidirectional search
• bringing the arm closer to the body.
• extending it towards the goal.

21
Configuration Sampling

• Consists of highly biasing the sampling towards
  the bent configuration of the elbow.

starts sampling the elbow flexion DOF with values
in the interval between 100 and 90 of flexion
22
Configuration Sampling
the x-component of the shoulder swing was
sampled between 50 and 100 degrees,
choose to sample higher arm postures.
23
Inverse Kinematics

• IK allows the user to define goal arm postures
  for the planner on-line, by simply selecting goal
  hand positions in the workspace.
• Analytical IK formulation
• Swivel angle

24
Inverse Kinematics
Swivel angle ?
Not valid
No
IK slover
? ?
? reach to Max Or min
Valid
Yes
Successfully
Failure
25
Result
26
Conclusion

• A new approach for motion editing based on
  planning motions in synchronization with recorded
  motion sequences.
• Our method is able to solve arbitrary
  spatio-temporal constraints among obstacles and
  takes into account dynamic environments.