Elbows in Virtual Environments Without Explicit Tracking

1
Elbows in Virtual Environments Without Explicit
Tracking

• (A Work in Progress)
• Chris Oates
• COMP 259 final project
• May 4, 2004

2
Background

• Want user to be able to see own arm in VE
• Adding trackers expensive, more wires
• Most trackers not very good
• See my presentation on Articulated Body Dynamics
• http//www.cs.unc.edu/oates/comp259/project/lectu
  re.ppt
  

3
Original Idea

• Treat head-to-hand chain as single articulated
  body chain, solve inverse problem for intervening
  joint angles
  
• Critical Error I cant do this because the act
  of nodding your head would require the force of
  moving the rest of the body, not just the head.
  Obviously wrong

4
More Hurdles

• Use 3rd party physics library?
• ODE Looks nice. But, Would not compile, need to
  get GMake for Windows, may have other issues
  since its Open Source/Linux-y and doesnt seem
  to like Windows/MSVC
• Wild Magic Now has Physics Engine
• BUT More a set of helper classes, does not
  appear to have constraint solvers, all solutions
  still done in application code
  
• Havok Commercial, , No Educational License
  program
  
• Running out of time, needed to get something
  working quickly

5
First Solution

• Not a research solution, just common-sense.
  Simplify problem into something easily solvable
  
• Use plain old IK, closed-form solution
• Treat shoulder to wrist as a 3-DOF arm. Two at
  shoulder, one at elbow
  
• Line from shoulder to wrist is reach axis
• If lengths of upper arm and forearm are known,
  this is just solving for the angles of a triangle
  with known sides!

6
First Test

• External 3-DOF Robot Arm in environment
• Base fixed to world point
• End-Effector always tries to reach object user
  can position in environment
  
• Always position elbow vertically above reach axis

7
Applying to Users Arm

• More common sense restrictions to simplify
  problem
  
• Make elbow always below reach axis instead of
  above
  
• Transform from head HiBall to shoulder
• Transform from hand HiBall to wrist

8
Head to Shoulder

• Assume rigid connection from head HiBall to base
  of skull
  
• Hang torso vertically from base of skull
• Let torso rotate with yaw of head, but remove
  pitch and roll
  
• Revisit the yaw issue in a bit

9
Hand to Wrist

• Because user holds a joystick in hand, can assume
  rigid link from hand HiBall to wrist
  
• No restrictions on rotations like for torso. If
  transform is correct, sphere on pointer will not
  move as hand is rotated

10
What Worked

• It looks okay
• Because user is actually moving trackers, we can
  assume continuity of motion, dont need to
  interpolate, just solve
  
• Users are willing to put up with a certain amount
  of discrepancy between visuals and reality
  
• They assume visuals are correct vs proprioceptive
  sense
  
• See Sharifs and Eric Burns work

11
What isnt Right

• Elbow not always below
• Arm reaching across body makes elbow pass through
  torso
  
• Shoulder does not follow head rotation
• Look over right shoulder (in real world) and
  model thinks shoulder is behind back
  
• Causes wrist to separate in model
• Also, can move shoulder forward to reach farther
• Also causes wrist separation

12
Minor Annoyances

• People are different sizes. Right now everything
  is coded to my arm lengths, shoulder width, etc
  
• Easy to change mathematical representation, but
  3D models dont always look right if just scaled
  
• Bending neck/waist to look at feet causes real
  problems, but not common enough to be big hassle
  
• Can cause tracker to lose view of ceiling, so
  discouraged

13
Can be fixed?

• Shoulder Assume shoulder is attached as before,
  unless cannot reach wrist position, then rotate
  shoulder to move it closer
  
• More trig, should be straightforward
• Elbow ???

14
Future Try a different solution

• Set up two articulated chains, add constraint
  force to keep roots in same location
  
• Will probably need additional constraints because
  general problem is underconstrained, likely to
  result in bad solutions
  
• Get ODE to compile and try it
• Compare to existing solution
• Perhaps a hybrid will work best

15
Demo