RO47005 - Planning and Decision Making

Delft University of Technology

January 25th , 2022 Name:
Exam questions Student number:

• This is a closed-book examination. You are only allowed to use one A4 paper with
handwritten equations, with no text other than sparse section headers.

• No calculators or other devices are permitted, nor needed.

• The examination time is 3 hours. (9:00 - 12:00)

• This exam consists of 5 questions.

• In total you can earn 40 points.

• To be able to receive all the points for a question you also have to add a solid argu-
mentation/computation.

• For each question an answer box is given. Make sure that your final answer fits within
the answer box. Only text within the answer boxes will be corrected.

• If you want to change your answer to a question you may use the answer boxes in
the last page. These will only be read if clearly specified in the main answer boxes.
They are intended for large changes/corrections. The space is the main boxes should
be enough to properly answer the questions.

• If you are unsure about a question, you may skip it and go to the next one. To come
back later.

• If you are following the Vehicle Engineering legacy course ”Robot Motion Planning &
Control”, you do not have to complete Question 5. This does NOT apply to those
following ”Planning & Decision-Making RO47005”!!

• Good luck!!!!
RO47005 Exam - Page 2 of 19 25-01-2022

1. (4 points) Consider a triangular robot, as shown in the figure below, that can only
translate on the plane. Consider the planar workspace of the figure, with obstacles in
gray. For this robot, both the workspace and the configuration space are planar, i.e.,
R2 .
(a) (2 points) Draw on top of the provided figure the configuration space obstacles for
this robot and environment.

The image below is provided to you as back-up in case you want to correct your answer.
If you use it, make it explicit which of the two we should correct, e.g., by crossing out
the wrong one.

Page 2
RO47005 Exam - Page 3 of 19 25-01-2022

(b) (1 point) Given start and goal configurations for this robot, describe the most suit-
able method to compute the collision-free path of shortest length between configu-
rations. Justify your choice.

(c) (1 point) What is a disadvantage of the proposed method?

Hint: think of what could be a problem if applied in real life.

Page 3
RO47005 Exam - Page 4 of 19 25-01-2022

2. (10 points) Consider the flying system in the following figure, which is formed by a
quadrotor and a rigidly attached, suspended arm. The quadrotor is actuated by four
independent rotors (each generating a thrust Ti ) and the arm is actuated with the two
indicated degrees of freedom, i.e., length l ∈ [lmin , lmax ] and angle θ without limits.

T4 T1

T3 T2

(a) (0,5 points) For this system, define and describe the workspace and the configura-
tion space.

We want to compute a collision-free path for this system from an initial to a final
configuration in an environment with obstacles. To do this, we will employ the
Rapidly Exploring Random Trees (RRT) method.
(b) (1 point) Is RRT a good choice for this system? Justify your answer.

(c) (0,5 points) Describe the sampling function to be used.

Page 4
RO47005 Exam - Page 5 of 19 25-01-2022

(d) (1 point) Describe the distance metric to be used.

(e) (2 points) Describe an appropriate steering function to be used.

(f) (1 point) Provide a detailed description of all the steps of the method employing
the defined functions.

Page 5
RO47005 Exam - Page 6 of 19 25-01-2022

(g) (1 point) What are the theoretical guarantees of the described method? Justify.

(h) (1,5 points) For this system, planning in configuration space can be cumbersome.
Is it correct to plan a path in some other space of lower dimensionality that the
configuration space? Justify your answer.

(i) (1,5 points) Consider now an environment with both static and moving obstacles.
Would you employ the derived RRT method, Velocity Obstacles or a combination
thereof? Explain in detail what would you employ, how and why.

Page 6
RO47005 Exam - Page 7 of 19 25-01-2022

3. (6 points) Consider a group of N robots like the one of question 2.

(a) (1 point) Describe the joint configuration space C. Justify your answer.

(b) (1 point) Give an advantage and a disadvantage of planning in the joint configura-
tion space.

(c) (2 points) Given a set of workspace obstacles O ⊂ W ⊂ R3 and the same team of
robots, describe the set of obstacles in the joint configuration space Oc ⊂ C.
Hint: in your answer you may denote by V(q) the volume occupied by a robot at
configuration q.

Page 7
RO47005 Exam - Page 8 of 19 25-01-2022

(d) (2 points) Consider the unlabeled problem, where we are given:

- the start configuration of all N robots, denoted by [s1 , . . . , sN ]. Denote the start
positions by [p1 , . . . , pN ] ∈ R3N ;
- a set of N goal locations for the robots, denoted by [g1 , . . . , gN ] ∈ R3N ;
- a set of obstacles in the workspace O ⊂ W .
Describe a method to plan collision-free paths for all the robots to reach the set of
goal locations while minimizing the total travelled distance (sum of the path length
for all robots).
Hint: consider the problem to be centralized and employ the joint configuration space
to plan collision-free paths.
Tip: if you do not have the answer, you may answer with a sub-optimal (approxi-
mate) method to get partial points. Justify your answer.

Page 8
RO47005 Exam - Page 9 of 19 25-01-2022

4. (10 points) Consider the following linear MPC problem with prediction horizon N
N
X −1
min xTk Qxk + uTk Ruk (1a)
u
k=0
s.t. xk+1 = Axk + Buk , k = 0, ..., N − 1 (1b)
x0 = xinit (1c)
Hxk ≤ g, k = 1, ..., N (1d)

in which xk ∈ Rn represents the state of the robot and uk ∈ Rm represents the control
command. The linear state constraints (1d) limit the state using H ∈ Rh×n and g ∈ Rh .

Hint: You can use the following notation (subscript ”N ”) for stacking a matrix M or
vector v as    
M v
 M 
v 
 
MN := , vN :=  ..  .
 
 ..
 .  .
M v
(a) (2 points) All predicted states of the MPC problem can equivalently be represented
as the linear function
   
x1 u0
x = Ψx0 + Φu, x =  ...  , u =  ...  , (2)
   
xN uN −1

of the initial state x0 and the inputs u. Derive Ψ and Φ such that this expression
holds. You can start with N = 1, N = 2 and then generalize to N ≥ 2, to finally
obtain the requested Ψ and Φ.

Page 9
RO47005 Exam - Page 10 of 19 25-01-2022

(b) (2 points) Show that, using expression (2), the constraints (1d) can equivalently be
written as
H̄u ≤ ḡ(x0 ) (3)
and give H̄ and ḡ(x0 ).

Page 10
RO47005 Exam - Page 11 of 19 25-01-2022

(c) (1 point) Consider Problem (1). We are given both a terminal set around the origin
X ∈ Rn and weight matrix P for a terminal cost. Formulate the MPC problem
that extends (1) with the terminal set and terminal cost.

(d) (2 points) What properties do the terminal set X and terminal weight matrix P
need to qualify as a valid terminal set and cost, respectively?

Page 11
RO47005 Exam - Page 12 of 19 25-01-2022

(e) (3 points) We assume now that MPC (1) is a motion planning problem with col-
lision avoidance as in Fig. 1. The robot state x = (x, y) describes its 2D position
and its collision region is modelled by a disc with radius rr . Our robot needs to
evade a single disc-shaped static obstacle that is represented by its center position
xo = (xo , yo ) and a radius ro .

Assuming that we know the difference vector (see Fig. 1),

   
x x
d= o − , (4)
yo y

derive a linear collision avoidance constraint hT x ≤ g that prevents collisions be-

tween the robot and the static obstacle. In your answer, h and g may contain the
state x, difference vector d and/or norm of the difference vector ||d||.

Obstacle

( , )

Robot

( , )

Figure 1: Schematic of the workspace in Question 4e

Page 12
RO47005 Exam - Page 13 of 19 25-01-2022

Page 13
RO47005 Exam - Page 14 of 19 25-01-2022

5. (10 points) Consider the drone in Fig.2. Its task is to take pictures of the flowers in a
field of tulips. The drone measures its 3D-location (x, y, z) relative to the plant. It has
also access to the angle between its camera heading and the tulip φ ∈ [−π, π]. To take
a picture of the plant, this angle needs to fall within a cone of [−π/6, π/6], but first
the drone needs to be closer than a distance d to the plant. Pictures can only be taken
during daylight, that is the light intensity l ∈ [0, 50000]% measured by the camera’s
sensor must be higher than a certain threshold τ . Finally, the behavior of the robot
should be reactive to perturbations such as the wind that may affect its state.
You are provided with the behavior tree in Figure 2 to complete the picture-taking task:

Figure 2: Initial sequence of actions

The only actions available to the drone are:

• Approach Plant, Orient Camera, and Take Picture

The drone can only sense:

• Angle between camera heading and target plant φ

• Light intensity l
• Own 3D-location (x, y, z)

Page 14
RO47005 Exam - Page 15 of 19 25-01-2022

(a) (1.5 points) Argue why the behavior tree in Fig.2 is poorly designed.

Page 15
RO47005 Exam - Page 16 of 19 25-01-2022

(b) (3.5 points) Design a new BT that correctly completes the task. The robot should
not take any action unless the light intensity is sufficient to take pictures, and the
BT should return SUCCESS only when the picture is taken (2pt). In addition,
the BT shall use explicit conditions for the actions as appropriate (1.5pt). Use
additional control nodes and conditions based on the available information only.
Draw the updated BT.

Page 16
RO47005 Exam - Page 17 of 19 25-01-2022

(c) (1.5 points) Define the return status of the three conditions you added at the pre-
vious step, as a function of the states x, y, z, φ, l.

(d) (3.5 points) A gust of wind instantaneously displaces the drone while it is executing
its behavior and the Take Picture node is RUNNING. As a result, the drone state
is (x = 10, y = 4, z = 3, φ = π/4). If the distance d = 11m, indicate which leaf
nodes are ticked in the next execution cycle of the BT that you have designed, and
what they return (2pt). Indicate how many seconds it will take to the drone to take
the picture, considering that it flies at a speed of 2m/s, the camera orientation and
taking the picture take 0.5s each, and the BT execution time is negligible (1.5pt).

Page 17
RO47005 Exam - Page 18 of 19 25-01-2022

Extra space for corrections - only read if clearly specified in the main answer boxes.

Page 18
RO47005 Exam - Page 19 of 19 25-01-2022

Extra space for corrections - only read if clearly specified in the main answer boxes.

Page 19