## Description

## Software

1. Read through the python tutorial here. You will need to use Numpy to complete parts of this

assignment. Some other resources are also available:

• Numpy tutorial from Stanford’s CS231n (link)

• Numpy for matlab users (link)

2. Install matplotlib. First do sudo apt-get install python-pip. Then do pip install matplotlib.

Vectors

1. (10 points) Given the following vectors:

p =

5

−1

1

, q =

−4

−2

7

, r =

−1

−2

3

Determine the following by hand (remember to show your work):

a. p + 2q

b. p · r and r · p where “·” denotes the dot product

c. q × r and r × q where “×” denotes the cross product

d. ||p||, and ||q|| where || · || denotes the Euclidean norm of a vector

1

向量相乘不需遵守矩阵相乘的条件

e. distance between the tips of p and q

2. (10 points) Find all k such that p =

−2

1

−k

and q =

2

−3k

−k

are orthogonal by hand.

Matrices

3. (5 points) Partition the following matrix into submatrices (i.e. find W, X, Y, and Z) by hand:

1 2 3 4

5 6 7 8

9 10 11 12

13 14 15 16

=

”

W X

Y Z #

where W ∈ R2×1 and Z ∈ R2×3

.

4. (10 points) Perform the following matrix multiplication by hand:

”

3 −1 −3

−1 0 2 #

0 1 2

−1 0 −1

0 −2 −1

2

1

−2

5. (12 points) Solve the following systems of equations using numpy. Recall that there can be a unique

solution, no solution, or infinitely many solutions.

a)

0 0 −1

4 1 1

−2 2 1

x =

3

1

1

b)

0 −2 6

−4 −2 −2

2 1 1

x =

1

−2

0

c) ”

2 −2

−4 3 #

x =

”

3

−2

#

6. (14 points) Given the following matrices:

A =

”

1 2

3 −1

#

, B =

”

−2 −2

4 −3

#

Use numpy to calculate the following.

a. A + 2B

b. AB and BA

c. A

T

, transpose of A

2

d. B

2

e. A

TB

T and (AB)

T

f. det(A), determinant of A

g. B

−1

, inverse of B

Rotation

7. (10 points) A rotation matrix R is defined by the following sequence of basic rotations:

i. A rotation of π/3 about the z-axis

ii. A rotation of π/4 about the new y-axis

iii. A rotation of π about the new z-axis

Compute the rotation matrix R.

8. (20 points) You would like to point a mobile robot so that it is looking at a target point. The camera

of the robot is aligned with x in the robot’s coordinate frame. The robot is assumed to be on flat

ground (z = 0). The robot’s pose in the world frame is:

T

0

r =

√

2

2

√

2

2

0 1.7

−

√

2

2

√

2

2

0 2.1

0 0 1 0

0 0 0 1

a. The target point is at p = [0.1, −1.4, 0]

T

in the world frame.

Assuming the robot stays at its

current position, calculate the vector v in the world frame that the camera should align with.

b. Use v to calculate the desired pose of the robot, again assume the robot does not change position.

The robot must remain upright on the ground; i.e. it’s z axis must not change.

c. Prove that the rotation component of the pose meets all conditions for being a valid rotation

matrix.

Least Squares

9. (30 points) You are testing one of the joints of a new robot arm and you notice there is some error when moving to a target position. To investigate, you command the joint to move through a

series of positions and measure where the joint moves for each command. The recorded data is

in calibration.txt, where the first column is the commanded position and the second column

is the measured position. You see that there is significant error in the commanded vs. measured

positions. Note: You are strongly advised to consult the linear algebra book, section 13.1 for this

problem, especially part (c).

a. (10 points) Use the psuedo-inverse to perform a least-squares fit of a line to the data. Produce

a plot showing the data as blue xs and the line you computed in red and include it in the pdf.

3

Include the parameters of the line and the sum of squared error of the fit in your pdf. Submit

your code as leastsquares.py to Gradescope. When we run the code, it should produce the plot

and print out the parameter values and the sum of squared errors.

b. (5 points) Is this least-squares problem underdetermined or overdetermined? Explain your answer.

c. (15 points) You notice that the error is quite high using a linear fit and you notice that the error

seems lower in the range [-0.5, 0.5] than elsewhere. Perform a piece-wise linear least-squares fit

to the data using -0.5 and 0.5 as the knotpoints. You should not perform perform multiple leastsquares fits to different parts of the data, i.e. you should set up the whole problem as Ax = b and

solve for all the parameters simultaneously.

Produce a plot in the same format as in (a) and include it in your pdf along with the parameters

you computed and the sum of squared errors. Use the parameters you computed in (c) to predict

what the measured position will be when you apply the command 0.68.

Include the prediction

in your pdf. Submit your code as pwleastsquares.py to Gradescope. When we run the code, it

should produce the plot and print out the parameter values, the sum of squared errors, and the

prediction for 0.68.

4