Professional Documents
Culture Documents
CPG Control of A Tensegrity Morphing Structure For Bio Mimetic Applications by Bliss, Smith, Iwasaki
CPG Control of A Tensegrity Morphing Structure For Bio Mimetic Applications by Bliss, Smith, Iwasaki
Abstract
The manta ray, Manta birostris, is an amazing creature, propelling itself through the water with the
elegant and complex flapping of its wings. This animal is of interest for morphing structures applications,
achieving outstanding efficiency and speed even with the enormous span of over five meters. This project
aims at integrating biomimetic control systems with morphing structures to harness what years of evolution
have created. Synthetic central pattern generators (CPG), the fundamental neural control mechanisms for
rhythmic motion in animals, are applied to actuation control of morphing tensegrity structures. Current
results illustrate successful integration of biomimetic control and structures to achieve efficient underwater
propulsion.
1 Introduction 2.1 Design and Synthesis of a CPG
1
Figure 1: Complete neural control architecture for rhythmic motion
is the sum of all j neurons, weighted by µij . These The assumptions made in this approach are valid
equations can also be written in vector form (3). because a periodic, oscillatory solution is the goal of
Pn
νi = ϕ(qi ) qi = F (s)ui ui = j=1 µij νj synthesizing a CPG.
(2)
ν = ϕI(q) q = F (s)M ν M (i, j) = µij (3) 2.1.4 Solving for the Connectivity Matrix
The problem now becomes finding the value for each
µij , commonly grouped as connectivity matrix M. With the problem linearized, the focus turns to
As can be seen, the problem is nonlinear, so the solving for the appropriate connections between the
method of harmonic balance is used to linearize the outputs and inputs of each neural element. First, we
problem. reduce the dynamic equations to
Bliss, T. K. 2
itself (self feedback), ties will have higher performance due to their low
weight and high actuation capabilities.
R= c s ci =αi cos(φi )
2 $ 3.1 Linear Algebra of Tensegrities
si = αi sin(φi ) Ω = (6)
−$ 2
$ = ω − 1/ω M = RΩR† K −1 To study tensegrity structures, a linear algebraic
approach is used3 , describing the structure as a sys-
where † represents the Moore-Penrose Inverse1 . tem of nodes and elements connecting those nodes.
This solution for M can now be used in numer- To begin, a nodal vector pi locates each node, νi ,
ically synthesizing a CPG using Matlab’s Simulink in Cartesian space and all nodes are compiled into
package. For example, Figure 2 shows the controlled vector p,
growth of rhythmic neuron activity in a six neuron
system with a 60 degree phase offset between each pix
neuron oscillating at 1 Hz. This neuron activity can pi = piy ∈ R3
now be used to control a structure, dictating desired piz (7)
actuator location or actuator force output.
pT = pT pT ... pT
1 2 n .
Let E represent the set of elements, where element
ei connects the j th and k th nodes: ei = {[νj , νk ], zi }.
The set E is comprised of sets Es and Eb , string
elements and bar elements respectively. The zi is an
element identifier such that
(
1, ei ∈ Es ;
zi =
−1, ei ∈ Eb . (8)
T
z = z1 z2 ... zn
A sparse connectivity matrix m is now con-
structed, and describes which nodes are connected
by which elements. Letting each row represent an
element, and each column a node, the matrix is filled
with 0, 1, and -1’s. A -1 means the element “leaves”
Figure 2: Synthesized CPG (three of six neurons the node, a 1 means the element “enters” the node,
plotted) and a 0 means the node is not associated with the
element. For example, if element 3 connects nodes 5
and 9, the third row of m would contain a -1 in the
fifth column, a 1 in the ninth column, and 0 else-
3 Tensegrity Analysis
where. As a further example, the matrix shown in
(9) is a suitable sparse connectivity matrix for figure
With the synthesis of a CPG complete, the focus
3.
switches to studying the structures the CPG will
control. We have chosen to investigate tensegrity
−1 1 0 0
structures, which are both statically and kinemati- −1 0 0 1
cally indeterminate. Simply stated, tensegrities are
0 −1 1 0
systems of bars held in compression by cables in ten- m= (9)
0 0 −1 1
sion. This state of self-stress, necessary for static
−1 0 1 0
equilibrium, makes the structure statically indeter- 0 −1 0 1
minate. These structures are also geometrically non-
This m matrix is expanded to the full connectivity
linear in that the position of the nodes are a func-
matrix M using the Kronecker product, M = m⊗I3 ,
tion of element lengths, which are again a function
where I3 is a 3x3 identity matrix. There is a corre-
of nodal position. Kinematic indeterminacy stems
sponding connectivity matrix CT = −ẑM, where
from the fact that elastic properties of the elements
keep the nodal positions from being described by x̃ = blockdiag{x1 , , x2 , ..., xn } ∈ Rmn×n , xi ∈ Rm
element lengths alone. These structures may seem x̂ = x̃ ⊗ I3 ∈ R3n×3n ,
overly complicated, but it is believed that tensegri- (10)
Bliss, T. K. 3
Figure 4, which describe a polyhedral cone, Figure
5, containing all possible configurations of the struc-
ture for the prescribed nodal location3, 4, 5 .
Bliss, T. K. 4
structure to produce the mechanics necessary for os-
cillatory propulsion. The beam slowly ramps up into
the bounded oscillations depicted in Figure 2. This
controlled growth in movement is what is often ob-
served in nature, as an animal accelerates up to an
optimal forward velocity. At steady state, the beam
oscillates as predicted, mimicking the motion of an
undulating fish tail.
Bliss, T. K. 5