Robotics I

June 10, 2009

Consider the planar PRP robot with n = 3 joints in the figure above. The world reference frame
RFw = (xw , y w , z w ) and the end-effector frame RFe = (xe , y e , z e ) are also shown.

• Assign the robot reference frames according to the Denavit-Hartenberg (DH) convention and
write down the associated table of parameters. Moreover, specify the (constant) transforma-
tion matrices w T0 , between the world frame and frame 0 of DH, and 3 Te , between frame 3
of DH and the end-effector frame.
• Based on the variables q defined in the Denavit-Hartenberg convention, compute the analyt-
ical Jacobian J (q) for a task involving only the end-effector position in the plane of motion,
and analyze its singular configurations. With the robot in a singular configuration q 0 , define
a set of base vectors for each of the following four linear subspaces:
       
R J (q 0 ) N J (q 0 ) R J T (q 0 ) N J T (q 0 ) .

• For a motion task of dimension m = 3 specified for the robot end-effector, consider the use
of a kinematic control law in the task space,
q̇ = J −1 (q) (ṙ d + K P (r d − f (q)) , (1)
where r ∈ R3 includes the position (in the plane) as well as the orientation of the end-effector
(i.e., the angle φ between the horizontal axis xw and the axis xe ) and f (q) is the direct
kinematic function associated to these task variables. Assume that the (positive definite)
gain matrix K P is chosen as diagonal, and let the joint velocities be bounded as |q̇i | ≤ Vi ,
with given values Vi > 0 (i = 1, 2, 3).

T
a) With the desired task velocity being ṙ d = 0 0 −1 , determine a joint configura-
tion q ∗ which is nonsingular for the task (and ‘matched’ to it, i.e., e = r d −f (q ∗ ) = 0 for
a suitable r d ) and is such that the desired task can never be realized without violating
one of the bounds on the joint velocities.

T
b) Let the robot initial configuration be q(0) = 1.2 π/2 1 , and let r d (0) =

T
T
1.5 1.5 −π/4 and ṙ d (0) = 0 1 0 be the specified task at time t = 0.
Define the numerical values of the diagonal matrix K P in the control law (1) so that the
initial task error e(0) will be reduced as fast as possible without violating the following
bounds on the joint velocities: V1 = 5 [m/s], V2 = π [rad/s], and V3 = 4 [m/s].

[180 minutes; open books]

 Solution
June 10, 2009

A possible assignment of the Denavit-Hartenberg frames is shown in the figure below, together
with the associated table of parameters.

i αi ai di θi
π
1 0 q1 0
2
π
2 0 0 q2
2
3 0 0 q3 0

From this, it is easy to obtain the general expression of the direct kinematics for this robot:
 0
R3 (q) 0 p3 (q)

0
T 3 (q) = = 0A1 (q1 ) 1A2 (q2 ) 2A3 (q3 )
0T 1
cos q2 0 sin q2 q3 sin q2
 
 0 −1 0 0 
=  .
 
 sin q2 0 − cos q2 q1 − q3 cos q2 
0 0 0 1

Use of the additional transformations between the frames defined in the problem text leads to

1 0 0 0 0 1 0 0
   
 0 0 1 0   0 0 1 0 
w
T e (q) = w T 0 0 T 3 (q)3 T e = 
0
 T 3 (q) 
  
 0 −1 0 0 

 1 0 0 0 
0 0 0 1 0 0 0 1
sin q2 cos q2 0 q3 sin q2
 
 − cos q sin q2 0 q1 − q3 cos q2 
 w w

2 Re (q) pe (q)
=  = ,
 
 0 0 1 0  0T 1
0 0 0 1

2
from which one can obtain the kinematic functions of interest (which could be derived also by direct
inspection, once the joint variables have been defined according to the Denavit-Hartenberg con-
vention). To this end, note that the final rotation matrix w Re (q) takes the form of an elementary
rotation matrix by an angle φ = q2 − π/2 around the world axis z w . In fact, it is
  π
π
 
cos q2 − − sin q2 − 0
 
cos φ − sin φ 0 2 2 sin q2 cos q2 0
w
 
 sin φ cos φ 0  =  π
π
  − cos q2 sin q2 0 
0=  = Re (q2 ).
 
 sin q2 − cos( q2 −
0 0 1 2 2 0 0 1
0 0 1

For the (first) task involving only the end-effector position on the plane, it is
! !
w
px q3 sin q2
r 1 = f 1 (q) = w = ,
py q1 − q3 cos q2

and the analytical (2 × 3) Jacobian matrix is

!
∂f 1 (q) 0 q3 cos q2 sin q2
J 1 (q) = = .
∂q 1 q3 sin q2 − cos q2

Analyzing the three minors of J 1 (q), this matrix looses rank if and only if sin q2 = 0 and q3 = 0, i.e.,
when the third robot link is oriented along the vertical direction and the third joint is completely
retracted. In such a configuration, the Jacobian becomes
!
0 0 0
J 1 (q 0 ) = ,
1 0 ∓1

where the upper sign corresponds to the case q2 = 0 and the lower sign to the case q2 = π. The
four linear subspaces indicated in the text are spanned by the following basis vectors:
     
   0 −1     1 
N J (q 0 ) =  1  ,  0  R J T (q 0 ) =  0  in R3 ,
0 ∓1 ∓1
   
   
0 1
   
R J (q 0 ) = N J T (q 0 ) = in R2 .
1 0

For the end-effector planar positioning and orientation task (of dimension m = 3), it is
w   
px q3 sin q2
r = f (q) =  w py  =  q1 − q3 cos q2  . (2)
   

φ q2 − π/2

The analytical (3 × 3) Jacobian matrix associated to this task,

 
0 q3 cos q2 sin q2
∂f (q) 
J (q) = =  1 q3 sin q2 − cos q2  ,

∂q
0 1 0

is singular if and only if sin q2 = 0.

3
 Under the condition of question a), namely with sin q2∗ 6= 0, it is
cos q2∗ sin q2∗ −q3∗ q3∗
    
0
1  1 
q̇ = J −1 (q ∗ )ṙ d =  0 0 sin q2∗  0  = − sin q2∗  .
  
sin q2∗ sin q ∗ 
2
1 0 −q3∗ cos q2∗ −1 q3∗ cos q2∗
It is then easy to see that, by sufficiently extending the prismatic joint 3, the robot will violate the
velocity bound at the first joint for any assigned value V1 > 0. More specifically, it is
q3∗ > V1 · | sin q2∗ | > 0 ⇒ |q̇1 | > V1 .

Under the condition of question b), the robot is not in a singularity at the initial time t = 0.
Thus, using the problem data and eq. (2), the initial control velocity can be computed as
 
q̇(0) = J −1 (q(0)) ṙ d (0) + K P (r d (0) − f (q(0))
      π 

0 1 −1 0 0.5 Kr1 1 + 0.3 Kr2 + Kr3
4
  π

=  0 0 1 
    
1 +  0.3 Kr2  =  − K ,

   r
  π   4 3 
1 0 0 0 − Kr3
4 0.5 Kr1
having set K P = diag{Kr1 , Kr2 , Kr3 }. From these expressions, one can directly choose two out of
the three control gains:
|q̇3 (0)| ≤ V3 ⇒ 0.5 Kr1 ≤ V3 = 4 ⇒ Kr1 = 8 > 0,
π
|q̇2 (0)| ≤ V2 ⇒ Kr ≤ V2 = π ⇒ Kr3 = 4 > 0.
4 3
Finally, using this definition, also the remaining gain is chosen:
π 10
|q̇1 (0)| ≤ V1 ⇒ 1 + 0.3 Kr2 + Kr = 1 + 0.3 Kr2 + π ≤ V1 = 5 ⇒ Kr2 = (4 − π) > 0.
4 3 3
With the selected gains, all joint velocities will saturate at time t = 0 (the second joint velocity
being at its negative limit −V2 = −π) and, as a result, the fastest decrease of the initial task error
e(0) = r d (0) − f (q(0)) will be realized (with the task error converging anyway exponentially to
zero, in a decoupled way for each task component). The situation at time t = 0 is depicted in the
following figure, where the desired initial robot configuration is the lightly shaded one.

∗∗∗∗∗