EECS4421Z: Introduction to Robotics

Sample exam questions

1. What is the forward kinematics problem for a robotic arm?

Solution: Given the joint variables and dimensions of the links, what is the position and orientation of
the end effector.

2. What is the inverse kinematics problem for a robotic arm?

Solution: Given the dimensions of the links and the position and orientation of the end effector, what are
the values of the joint variables.

3. Consider the arm shown in Figure 1.

Figure 1: In this figure, θ1 and θ2 are shown at 0◦ .

(a) Solve the forward kinematics problem.

Solution: Note that the frame of the end effector is not fully specified (only its position is specified);
thus, any sequence of transformations that yields the end effector position in frame {0} is acceptable.
One such sequence is:

T30 = Rz,θ1 Dz,a Rx,−90 Rz,θ2 Dx,d3 (1)

 
cos θ1 cos θ2 − sin θ1 cos θ1 sin θ2 d3 cos θ1 cos θ2
 cos θ2 sin θ1 cos θ1 sin θ1 sin θ2 d3 cos θ2 sin θ1 
= 
 − sin θ2
 (2)
0 cos θ2 a − d3 sin θ2 
0 0 0 1

The final matrix is not required, but is useful to double check the inverse kinematics solution.

Page 1 of 6
(b) Given the location of the end effector (x, y, z)T , solve the inverse kinematics problem (assume d3 ≥ 0).

Solution: Looking at the robot from the top down yields θ1 :

θ1 = atan2(y, x) (3)
This can be verified by examining the x and y coordinates of the position from the forward kinematics
solution ( xy = dd33 cos
cos θ2 sin θ1
θ1 cos θ2
sin θ1
= cos θ1
= tan θ1 ).
p
θ2 can be found from the right angle triangle with hypotenuse d3 , base x2 + y 2 , and side a − z:
p
θ2 = atan2(a − z, x2 + y 2 ) (4)
A second solution exists for this robot:

θ1 = atan2(y, x) + 180 (5)

p
θ2 = −180 − atan2(a − z, x2 + y 2 ) (6)

Page 2 of 6
4. Consider the arm shown in Figure 2.

Figure 2: In this figure, θ1 and θ2 are shown at 0◦ .

(a) Solve the forward kinematics problem.

(b) Given the location of the end effector (x, y, z)T , solve the inverse kinematics problem (assume d3 ≥ 0).

5. Complete or add frames suitable for the Denavit-Hartenberg (DH) convention for the robot shown in Question
3, and give the table of DH parameters.

6. Complete or add frames suitable for the Denavit-Hartenberg (DH) convention for the robot shown in Question
4, and give the table of DH parameters.

7. Consider the differential drive mobile robot (shown below) with wheel spacing, L, left wheel ground velocity,
vL , right wheel ground velocity, vR , angular velocity around the ICC, ω, and distance between the robot origin
and the ICC, R.
vL vR

ω ICC

R
L

(a) Derive the equations for vL and vR in terms of ω, R, and L.

Page 3 of 6
 Solution: The relationship between linear velocity v and angular velocity ω for a point rotation
about the ICC at a distance r from the ICC is given by ω = v/r. All that is required is to compute
the appropriate value of r for the left and right wheels:

L
vL = ω(R − ) (7)
2
L
vR = ω(R + ) (8)
2

(b) A robotics student proposes a course project where they wish to modify an existing differential drive
robot to make its left wheel steerable. What is your advice to the student? Justify your answer based on
the kinematics of the differential drive.

Solution: This is a bad idea. The ICC for the robot will be located at the point where the left wheel
contacts the ground. If the left wheel is not steered in a direction exactly equal to that of the right wheel
then the robot will want to spin around the contact point of the left wheel and the ground.

8. (a) Sketch ellipses on the axes provided corresponding to the given covariance matrices. The exact dimen-
sions of the ellipses are not important, but the sketches should be reasonable approximations of the true
answers.
  Y
1 0
i.
0 1

  Y
3.25 1.299
ii.
1.299 1.75

Page 4 of 6
 Hint: the eigenvalues for the last matrix are 1 and 4.
(b) Suppose you have two measurements z1 = 10 and z2 = 20 of the distance between a robot and an object.
Assuming z1 ∼ N (0, σ12 ) and z2 ∼ N (0, σ22 ) what is the best estimate of the distance if:
i. σ1 = σ2

Solution: The measurement variances are equal; thus the best estimate is simply the average of
z1 and z2 (i.e, 15).

ii. 3σ1 = σ2

Solution: The measurement variances are unequal; choose weights:

σ12 σ12 1
w2 = 2 2
= 2 2
= (9)
σ1 + σ2 σ1 + 9σ1 10
9
w1 = 1 − w2 = (10)
10
and compute the weighted average:
9 1
µ = w1 z1 + w2 z2 = 10 + 20 = 11 (11)
10 10

9. What quantity does the velocity and odometry motion models compute?

Solution: p(xt |ut , xt−1 ) the probability that the robot ends up at pose xt given that it started from pose
xt−1 and moved with control vector ut .

10. Why is it more difficult to compute a motion model than it is to sample from a motion model?

Solution: A motion model solves an inverse kinematics problem (what control vector ût is required to
move the robot from pose xt−1 to pose xt ) and uses the solution of the inverse kinematics problem to
compute p(xt |ut , xt−1 ). Sampling from a motion model requires a solution to the forward kinematics
problem which is usually easier to solve than the inverse kinematics problem.

11. Suppose that a robot is equipped with a sensor for measuring range and bearing to a landmark, and assume that
the sensor can determine which landmark it is currently measuring. Assume that we are performing global
localization with an extended Kalman filter. When measuring a single landmark, the probability distribution
of the robot location is usually poorly approximated by a Gaussian distribution. When measuring two or
more landmarks, the probability distribution of the robot location is usually well approximated by a Gaussian

Page 5 of 6
distribution. Explain why (hint: consider the case where the position and heading of the robot are highly
uncertain).

Solution: Suppose that we know that the robot is somewhere to the left of the landmark and that its
heading is unknown. Suppose that the robot makes a measurement of a single landmark to obtain a
distance and bearing relative to the robot (shown as an arrow in the follow figures). One possible pose of
the robot relative to the landmark is shown below:

But there are many robot poses that could have generated the same measurement of the landmark:

Assuming perfect measurements, the probability distribution of the robot location is a half circle, which
is not Gaussian. With noisy measurements, the probability distribution looks approximately like a half
ring of non-zero width.
If you try to draw similar pictures where the robot has measurements of two landmarks that are not too
close to one another then you will find that for most arrangements of markers the probability distribution
of the robot location is not too far from Gaussian.

Page 6 of 6