Self-Collision Detection
and Prevention
for Humanoid Robots
James Kuffner et al.
presented by Jinsung Kwon
Self-Collision
Mobile Robots are
free of self-collisions
in most cases
Self-Collision
Ariticulated robots are typically at
high risk of self-collision
Objective
Develop efficient
geometric method
• detect and prevent
self-collision
• suitable for complex
articulated robots
H7 Humanoid
(31 Links)
Challenges
Large number
of distance
computations
in short time
N = 31
P = 435
Challenges
Single distance computation itself
is also very expensive
Strategies
Eliminate unnecessary pairs
from distance computation
Strategies
Eliminate unnecessary pairs
from distance computation
Strategies
Protective Hulls
approximation to the complicated geometry
Strategies
Protective Hulls
Implementation
Trajectory Sampling
: discretization of the trajectory into a
finite set of samples
Implementation
Velocity Bounds and
Collision-free Guarantees
No Collision if
xmax < dmin
during ∆t
xmax
dmin
with dx = J dq
|dq/dt| < (dq/dt)max
Implementation
Voronoi-clip for distance computation
• Running time depends on the geometric
complexity and posture changed
• Running relatively in constant time with high
coherency
• Limited to convex polyhedrons
Implementation
Control Strategy
0
1
2
3
1
2
3
Read joystick command
2
4
Final Posture
by Emergency
Stop
3
4
5
3
4
5
Calculate 3-step trajectory
Check new trajectory
for self-collision
6
Results
Results
Results
Comparison
Future Work
• Automatic selection of active
pairs for given joint angle ranges
• Alternative minimum distance
determination method allowing
non-convex protective hulls