RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
What’s Wrong with Collision Detectionin Multibody Dynamics Simulation?
Daniel Montrallo Flickinger Jedediyah Williams Jeffrey C Trinkle
CS Robotics Laboratory
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Multibody Dynamics Simulation
Multibody System Dynamics involves the modeling and simulation of systems ofinterconnected bodies
Why should I care? So what is wrong with it?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Why should I care?
Dynamic simulation is important ...
Ï for robotics and controls
Ï for designing complex machinery
Ï for virtual reality
Ï for the entertainment industry
Demand for high fidelity dynamic simulation is high as system complexityincreases under limited computational resources.
So what? I still don’t care Okay, that makes sense, but what is wrong with it?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
What’s wrong with collision detection in multibody dynamicssimulation
Ï Low fidelity
Ï Slow performance
Ï Tuning required (“magic numbers”)
Ï It’s slow
Why should I care? How do you solve it then?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
How we solve the problems with collision detection
P
1
23
4
ψ1n =ψ2n
Ï Formulate constraints that are geometrically accurate: the Polyhedral ExactGeometry formulation
Ï Devise efficient collision detection algorithms and complementarity systemsolvers
What are constraint formulations? Tell me more about PEG
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
What are constraint formulations?
M
ψ
p1
p2
Ï Gap distance φ is calculated for each pair ofgeometric features
Ï Unilateral constraint equations constructed foreach gap distance
What are the complementarity conditions? What is the Stewart-Trinkle formulation?
What is the Polyhedral Exact Geometry formulation? I don’t care, just show me some pretty graphs
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Complementarity Constraints
M(q, t)ν̇=λvp(q, q̇, t)+λapp, (1)
where λvp is the sum of velocity dependent forces, and λapp are the applied forces.Complementarity constraint:
0 ≤λl+1n ⊥ GT
nνl+1 +Ψ
ln
h+ ∂Ψl
n
∂t≥ 0, (2)
where λn are forces normal to contact surfaces,Ψn are gap functions, or distancesbetween active bodies and contact surfaces, h is the step size, ν are the velocitiesof the active bodies, and Gn is a normal contact wrench.
What is the Stewart-Trinkle formulation? What is the Polyhedral Exact Geometry formulation?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Stewart-Trinkle formulation
ψminimum
ε
t1
t2 t3
t4
Show me the Polyhedral Exact Geometry formulation So let’s see the equations
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Stewart-Trinkle formulation, mixed linear complementaritysystem
Equations of motion and complementarity equations (1) and (2) are formulatedas a mixed LCP: ∣∣∣∣ 0
ρl+1n
∣∣∣∣= ∣∣∣∣ M −Gn
GTn 0
∣∣∣∣ ∣∣∣∣νl+1
pl+1n
∣∣∣∣+ ∣∣∣∣−Mνl −plext
ψln/h
∣∣∣∣ , (3)
where M is the inertia matrix, ν is the generalized velocities, pn and pext arenormal and external impulsive forces, and h is the step size.Normal contact wrench:
Gnij =[
n̂ij
rij × n̂ij
], (4)
Show me the Polyhedral Exact Geometry formulation Jump right to the PEG equations
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Polyhedral Exact Geometry formulation
ε
t1
t2
t3
So let’s see the equations That’s cool, what are the results? What is the Stewart-Trinkle formulation again?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Polyhedral Exact Geometry formulation, mixed linearcomplementarity system
The Stewart-Trinkle forumation in (3) is altered to introduce a heuristic:∣∣∣∣∣∣0
ρl+1n
ρl+1a
∣∣∣∣∣∣=∣∣∣∣∣∣
M −Gn 0GT
n 0 E1
GTa 0 E2
∣∣∣∣∣∣∣∣∣∣∣∣∣νl+1
pl+1n
cl+1a
∣∣∣∣∣∣∣+∣∣∣∣∣∣−Mνl −pl
extΨl
n/h∆Ψa/h
∣∣∣∣∣∣ , (5)
where M is the mass matrix, and Gn and Ga are the normal and auxiliary contactwrenches, respectively. Multiple adjacent contacts are grouped, and thecomponents of Gn from (4) are split into Gn and Ga in (5). Gn contains the contactwrenches of the contacts with the minimum value of ψn for each group ofcontacts.
Yes, but what is this auxiliary contact wrench? What is the Stewart-Trinkle formulation again? Show me an example
Whatever, just show me the results
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Polyhedral Exact Geometry, auxiliary contact wrench
The remaining contact wrenches in each manifold are put into Ga. The auxiliarygap functions are defined as
Ψa =
Ψa1
...Ψans
whereΨaj =
Ψ1 −Ψ2...
Ψ1 −Ψns
. (6)
Wait, what? Show me an example That’s cool, what are the results?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Polyhedral Exact Geometry formulation, two constraintexample
M
ψ1ψ2
p1
p2
p3
max(ψ1,ψ2) ≥ 0 (7)
max(ψ1,ψ2) =ψ2 +max(0,ψ1 −ψ2). (8)
c = max(0,ψ1 −ψ2) (9)Complementarity constraints:
0 ≤ c− (ψ1 −ψ2) ⊥ c ≥ 0. (10)
0 ≤ max(ψ1,ψ2) ⊥λj ≥ 0, (11)
Which gives
0 ≤c+ψ2 ⊥λ1 ≥ 0 (12)
0 ≤c+ψ2 ⊥λ2 ≥ 0. (13)
Very interesting, do you have a general example? Forget I even asked, just tell me the results
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Polyhedral Exact Geometry formulation, general exampleExpanding the constraint in (10), a general system with m facets in contact is realized,
0 ≤ c2 −ψ2 +ψ1 ⊥ c2 ≥ 0 (14)
0 ≤ c3 −ψ3 + c2 +ψ1 ⊥ c3 ≥ 0
.
.
.
0 ≤ cm −ψm + cm +cm−1 + ...+ c2 +ψ1 ⊥ cm ≥ 0
0 ≤ d1 +ψ1 ⊥ d1 ≥ 0
0 ≤ d2 +ψ2 ⊥ d2 ≥ 0
.
.
.
0 ≤ dm +ψm ⊥ dm ≥ 0
0 ≤ d1 + (c2 + c3 + ...+ cm−1 + cm)+ψ1 ⊥λ1 ≥ 0
0 ≤ d2 + (c2 + c3 + ...+ cm−1 + cm)+ψ2 ⊥λ2 ≥ 0
.
.
.
0 ≤ dm + (c2 + c3 + ...+ cm−1 + cm)+ψm ⊥λm ≥ 0
0 ≤ (c2 +c3 + ...+ cm−1 +cm)+ψ1,
whereci = max(0,ψ1 −ψi), i = 2, . . . ,m, (15)
and d are slack variables. Note that ψi := [ψn
]i .
Okay, so what are the results?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Results
Trajectory error results are compiled for three uncomplicated benchmarksimulations
Ï sawtooth particle simulation
Ï hills particle simulation
Ï slender rod 3D simulation
1.99 2 2.01 2.02 2.03 2.04 2.05
0.995
1
1.005
1.01
1.015
1.02
1.025
X (meters)
Y(m
eter
s)
PEGS-TA-P
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Sawtooth benchmark simulation
0 2 4 6 8 12 140
0.51
1.5
X (meters)
Y(m
eter
s)PEG
SâT
4 4.2 4.4 4.6 4.8 5 5.2
0.6
0.8
1
1.2
1.4
1.6
1.8
X (meters)
Y(m
eter
s)
PEGSâT
Okay, how about another benchmark example? Particle simulations are junk, why don’t you show me a real system?
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Hills benchmark simulation
−1 −0.50 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
X (meters)
Y(m
eter
s) PEGSâT
0.5 1 1.5 2 2.50.5
1
1.5
X (meters)
Y(m
eter
s)
Alright, that’s nice. Now show me a 3D example Thanks, but I’m done
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
Slender rod benchmark simulation
10−4 10−3 10−2 10−110−3
10−2
10−1
100
h (s)
med
ian
erro
r(m
)
PEGAPST
Cool beans
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
...
RensselaerPolytechnic
Institute
Daniel MontralloFlickinger,Jedediyah
Williams, Jeffrey CTrinkle
IntroductionBackground
The Problem
FormulationsGeometric Constraints
Stewart-Trinkle
Polyhedral Exact Geometry
ResultsSawtooth Benchmark
Hills Benchmark
Slender Rod Benchmark
...