Professional Documents
Culture Documents
contact/interaction between the robot and a purely geometric (rigid and frictionless) environment naturally constrains the end-effector motion in ideal conditions (robot and environment perfectly rigid, frictionless contact), one can define in the task space two sets of generalized directions selected so that
!!
!!
!!
end-effector motion is feasible in a set of k directions natural constraints (where the environment cannot react with forces/torques) on force and motion contact reaction forces/torques arise in a set of 6-k directions imposed by the task (where the environment bars any end-effector motion)
!!
these sets of directions are mutually orthogonal (and complementary, namely they cover the 6D task/Cartesian space) and are characterized by a suitable task frame RFt (typically attached to the robot end-effector) for general interaction tasks, position and orientation of the task frame will be time-varying the particular way task execution should be performed can be expressed in terms of artificial constraints that specify desired values (to be imposed by the control law) for the velocities, in the k directions feasible for motion, and for the forces, in the remaining 6-k directions feasible for contact reaction
2
!!
!!
Robotics 2
- example 1
RFt
y x
v = linear velocity ! = angular velocity F = force M = moment
task: slide the cube along the guide natural (geometric) constraints vy = v z = 0 6-k=4 !x = !z = 0 k=2 Fx = My = 0 artificial constraints (to be imposed by the control law) Fy = Fy,des = 0 Mx = Mx,des = 0, Mz = Mz,des = 0 Fz = Fz,des !y = !y,des = 0 vx = vx,des
3
6-k=4 k=2
Robotics 2
Selection of directions
z
- example 1
0& 0( ( 0( # vx & # vx & )% ( = T )% ( ( 0 $" y ' $" y ' ( 1( 0( '
RFt
y
"0 x $1 $ " F % $0 $M ' = $ 0 # & $ $0 $0 #
Robotics 2
0 0 0% 0 0 0 ' "! F " Fy % ' $ y% $F ' 1 0 0 ' Fz ' z $ ' $ ' ( = Y ( ' 0 1 0 $Mx ' $Mx ' ' $ ' $M ' 0 0 0 ' # Mz & # z& 0 0 1' &
TT Y = 0
constraint forces/torques cannot perform work
4
- example 2
task: turning a crank (free handle) natural constraints vx = v z = 0 !x = !y = 0 Fy = Mz = 0 artificial constraints Fx = Fx,des = 0, Fz = Fz,des = 0 Mx = Mx,des = 0, My = My,des = 0 vy = vy,des !z = !z,des = 0
"
x0
RF0
Rz(")
RFt(")
Robotics 2
Selection of directions
z RFt(") z0 x y RF0
example 2
0& 0( ( 0( # vy & #vy & * = T ( ) ) * %" ( 0( % "z ( $ ' $ z' ( 0( 1( '
"
#0 %1 y0 % 0 T # v & #R () ) 0 &% 0 = %0 ( % ( R T () ) '% 0 x0 $ " ' $ 0 % %0 %0 $ 0 0 1 0 0 0 0 0 0 1 0 0 0% 0 ' " Fx % " Fx % ' $ ' $F ' 0 ' Fz $ ' = Y (( ) ) $ z ' ) 0 ' $Mx ' $ Mx ' ' $ ' $M ' 1 ' #My & # y& 0' &
TT(") Y(") = 0
- example 3
x y
p = screw pitch
RFt
artificial constraints (abundant) Fx = Fx,des = 0, Fy = Fy,des = 0 Mx = Mx,des = 0, My = My,des = 0 vz = vz,des , !z = !z,des = (2#/p) vz,des Fz = Fz,des , Mz = Mz,des (Fz,des )
7
the screw proceeds along and around the z-axis, but not in an independent way! (1 dof) accordingly, Fz and Mz cannot be independent
Robotics 2
Selection of directions
example 3
2) p
* vz = T * vz
k=1
! p
RFt
z
"F % $M ' # &
6-k=5
the columns of T and Y do not necessarily coincide with Cartesian directions (columns of the identity matrix) generalized directions
"1 $0 $0 =$ $0 $0 $0 #
0 0 0 0 % " Fx % " Fx % $ ' 1 0! 0 0 ' $ F ' Fy ' 2) ' $ y ' $ 0 0 0 ( p ' * $M ' = Y * $M ' x x 0 1 0 0 ' $ ' $ ' M M y' $ y' 0 0 1 0 ' $ $M ' $M ' 0 0 0 1 ' # z & # z& &
8
Robotics 2
Frames of interest
y0 x0 yt
example 4
ye
xt xe
environment surface
"! task frame RFt used for an independent definition of the hybrid reference values (here: tvx,des [k=1] and tFy,des [M-k=1]) and for computing the errors driving the feedback control law "! sensor frame RFe (here = RF2) where the force eF = (eFx,eFy) is measured "! base frame RF0 in which the end-effector velocity is expressed (here, 0v = (0vx,0vy) of O2), when computed using robot Jacobian and joint velocities all quantities should then be expressed (rotated) in the same reference frame: the task frame!
Robotics 2 9
#v& k = T ( s ) ) s s " IR %" ( $ ' parameterizes "F % ! M$k " # IR = Y ( s ) ( ) $M ' # & parameterizes
TT(s) Y(s) = 0
the generalized directions contact forces/torques of the task frame depend in general on s (i.e., on the E-E pose in the previous first three examples, ! in the environment) and in! general, it is M=6
robot dynamics
Robotics ! 2
control objective: to impose the desired evolution to the parameters s of motion and to parameters $ of force s(t) sd(t), $(t) $d(t) control law is designed in two steps
1.! exact linearization and decoupling in the task frame by feedback closed-loop model
!!
2.! (linear) design of as and a$ so as to impose the desired dynamic behavior to the errors es = sd s and e$ = $d $
. Note: in simple cases, $ and s are just single components of F or M and of v or !; accordingly, Y and T will be simple 0/1 selection matrices
11
!
"1
linearizing and $ as ' "1 " Jq + S* q +g u = (BJ T " J Y )& ) + BJ Ts decoupling a % #( control law
"1 T
!
Robotics 2
!
!
!
!
Robotics 2
!
13
r L
!
s="
x0
= T # (s)J(q) " q s
pseudoinverses of tall matrices with full column rank, e.g., (TTT)-1TT (or weighted)
14
as
KI s + a$
!+
$d
s, s
!
!
$ s filtering of measures s
KD
KP
limit cases k=M (free motion): no force control loops, only motion ! k=0 (frozen robot end-effector): no motion control loops, only force
Robotics 2 15
. $ and s are just single components of F (or M) and of v (or !) Y and T are replaced by 0/1 selection matrices: and I-
Robotics 2 16
video
17
Sources of inconsistency
if the geometry of the environment at the contact is known with precision, task inconsistencies due to 1. and 2. on the measures of s and $ are automatically filtered out through the pseudo-inversion of the matrices T and Y
(can be reduced/eliminated by real time estimation processes driven by external sensors: vision, but also force!)
18
Robotics 2
[degrees]
Robotics 2
20
[degrees]
Robotics 2
21
differences are in the order of 7-8... but which one is correct? Better results are obtained with some kind of sensor fusion!
[degrees]
Robotics 2
22
[m]
this is the reconstructed contour of a cinema film reel (of radius = 17 cm)
Robotics 2
[m]
23
Normal force
regulated to 20 N during simultaneous motion and identification
peaks correspond to grooves on the surface contour
[N]
video
Robotics 2
26
!!
!!
car windshields with sharp edges and tolerances due to fabrication and excess of gluing material (PVB=Polyvinyl butyral) between glass layers robot end-effector follows a preprogrammed path, despite the small errors w.r.t. the nominal windshield profile, thanks to the passive compliance of the deburring work tool contact force between blades and work piece can be independently controlled by a pneumatic actuator in the work tool
!!
!! !!
the deburring robotic worktool contains in particular: two blades for cutting the exceeding plastic material (PVB), the first actuated, the second passively pushed by a spring a load cell for measuring the 1D applied force on-board control system
27
Robotics 2
physical mathematical
Robotics 2
for a stability analysis (based on a linear model and on root locus techniques) of a force control loop in a single direction, in the presence of multiple masses/springs as in this present case, see Eppinger & Seering, IEEE CSM, 1987 (material in the course web site)
28
force control (realized as external loop providing the reference to an internal position loop)