A Motion Planner for the Human Hand

1
A Motion Planner for the Human Hand

• Project by
• Qi-Xing Samir Menon

2
Motion Planning for the Human Hand
?i
?2
?1
?20
Find Parametrization Vector, T?1, ?2..
Generate Hand Skeleton
Define Configuration Space
User defines two poses Find Path Smoothen to
get Realistic Motion
Sample Configuration Space for Milestones
Collisions
Connect Adjacent Configurations
3
The Human Hand

• Motion is induced by the application of
  musculo-skeletal control
• We demonstrate planned motion of a human hand
• Simulated hand has a 20 degree of freedom
  skeleton
• Control is applied to the joint angles of the
  skeleton
• Planning takes place in 20-dimensional joint
  configuration space
• The planned path is executed in a simulated model
  of the human hand

4
The Hand Skeleton

• The human hand may be modeled using a 20 DoF
  skeleton parameterization
• Configuration of the human hand is represented by
  a 20 dimensional joint angle vector

Hand Space
5
Modeling the Hand

• Hand-space models an actual human hand
• The hand is represented by a mesh representation
  of a laser scanned hand
• The parameterization allows the emulation of a
  real hand

Hand Space Configuration
6
Configuration Space

• Hand motion is in 20 dimensional configuration
  space along the planned path

?i
?20
Disallowed Hand Config CSpace Obstacle
T1
Path of Motion
T2
?2
T3
T4
T5
Ti?1, ?2,, ?20 Represents a hand configuration
?1
?0
Milestones Sampled Hand Configuration
7
Uniform Sampling

• A uniformly random sampler

?i
?20
?2
?1
?0
8
Adaptive-Gaussian-Random Sampling

• An adaptive gaussian sampler

?i
?20
Gaussian Sample
?2
Adaptive Sample
Random Sample
?1
?0
9
Connecting Samples

• Obtain a roadmap in the form of a search graph
• Connect each sample to 10 closest samples and
  check for collision
• Reject connections with collisions

?i
?20
?2
?1
?0
10
Collision Detection Strategy
?i
?20
?2
Collision!!!
Path Added To Roadmap!
?1
?0
11
Planning Hand Motion

• Add start and goal configuration nodes to graph
• Search for a path in the graph

Resulting Path is Jerky due to imperfect
sampling!!
?i
?20
?2
Goal
?1
Start
?0
12
Planning Hand Motion (contd.)

• Video of jerky motion

13
Smoothing Motion
?i
?20
Smooth Path is obtained!!
?2
Goal
?1
Start
?0
14
Smoothing Motion (contd.)

• Video of smooth motion

15
Demo
Eg.4

• System demo

Eg.1
Eg.2
Eg.3
16
Results Sampling
17
Results Smoothing
Smoothed Path Length Smoothed Milestones Time (sec) Unsmoothed Path Length Unsmoothed Milestones Time (sec)
Eg.1 3.40 2 1.650 6.58 5 0.650
Eg.2 4.11 2 0.735 6.16 5 0.698
Eg.3 2.34 3 0.605 2.45 4 0.585
Eg.4 2.95 3 0.590 3.00 3 0.565
18
Discussion

• Smoothing the path greatly improves motion
  quality
• Adaptive Gaussian Sampling can drastically reduce
  the required samples but it also requires more
  precomputation
• Straight line motion in higher dimensional space
  produces better quality than curved or spline
  motion.

19
Future Work

• Areas for improvement
• The project may be extended to involve
• Control applied to muscular configuration space
• Improved skeleton that closely matches a real
  hand
• System dynamics such as inertia and damping