TENTA PART 2 FRÅGOR

Question 1

Please answer the following question about driver models
a) Describe the difference between rule-based and data-driven driver modeling (1p)

b) Give one positive and one negative aspect of each of these model types. (1p)

c) Describe the terms baseline and treatment in safety assessment (1p)

Answer 1

a) In rule-based modeling there are functions specified by the modeler (typically equations with specific parameters). Data-driven models learn behavior from data, such as neural networks, reinforcement learning etc.

b) These are possible pros and cons:
a. Rule-based models:

i. Pro: Typically a clear connection between mechanics and predicted behavior
ii. Pro: Typically possibly to understand what is happening by studying the
model and its parameters
iii. Con: Limited in flexibility and generalizability – typically narrow context for
each parameterization

b. Data-driven models:

i. Pro: Good flexibility in reproducing behavior
ii. Con: Mostly no link to human cognition.
iii. Con: Typically interpretability very limited (understanding why things happen
the way they do).

c) Baseline and treatment.

a. Baseline: Data without the assessed safety solution, serving as a benchmark for performance comparison.

b. Treatment: Application of a safety solution to prevent or reduce crash impact in a scenario.

Question 2

In the context of automated driving (SAE J3016), please define:

1) the operational design domain (1 point)
2) the dynamic driving task (1 point)
3) the limp mode (1 point)

Answer 2

1.Operational Design Domain (ODD): Defines the operating conditions for an automated system, commonly used in autonomous vehicles.

2.Dynamic driving task (DDT) All real-time functions needed to operate a vehicle in traffic within its specific design limits.

3. A vehicle security feature that limits speed or motion range when Level 4 or 5 ADS performs the DDT fallback.

Question 3

In the context of active safety,
1) What is an adaptation failure? (1 point)

2) Please, give at least two examples of adaptation failures (1 point).

3) What is driver impairment and how does it relate to adaptation failures (1 point)?

Answer 3

1) An adaptation failure occurs when a driver exceeds the safety zone boundary because of erroneous perception, overestimation, misunderstanding, or unexpected events.

2) Panic steering when a deer suddenly crosses the road. (The unexpected event may trigger some
unusual steering that may make the driver cross the safety zone boundaries)

3) Driver impairment includes distraction, intoxication, and fatigue. All these types of impairment may
increase the frequency and the severity of the consequences from adaptation failures

Question 3

Please answer the following questions related to crash databases.

A) What is a crash database, please mention the name of at least one national crash database (1 point)?

B) Please give at leasttwo examples of crash-related variables that may be present in crash database and at least two examples of crash-related variable that many not be present in crash databases (1 point)

C) Can information about injuries be available in crash databases? If so, what would be a standard way to measure injury severity, please explain how such a measure would rank different injury levels (1
point)?

D) What is an injury curve and what does it represent? Please provide a drawing of an injury curve as you answer this question (1 point)

Answer 3

Injury Curve: Shows the probability of injury severity levels based on crash factors, like impact speed. It illustrates how injury risk rises with speed and compares the vulnerability of cyclists, pedestrians, and motorcyclists to drivers.A. Crash databases include data about crashes collected by the police (and sometimes hospitals) after a crash has occurred. STRADA is the Swedish national database.

B.
a. Included:
i. Date of the crash
ii. Type of the vehicle involved in the crash

b. Not included
i. Glance behaviour of the road users involved in the crash
ii. Braking pressure at the time of the impact

C. Yes. The AIS: abbreviated injury scale is a standard.

D.Injury Curve: Shows the probability of injury severity levels based on crash factors, like impact speed. It illustrates how injury risk rises with speed and compares the vulnerability of cyclists, pedestrians, and motorcyclists to drivers

Question 3

c) Describe the terms fixation, saccade and smooth pursuit in eye-tracking (1p)

Answer 3

c) Fixation: When eye gaze is fixated (looking at) in a specific position
Saccades: when it moves to another position
Smooth pursuit: “Fixation” on moving objects

Question 4

This question is about cooperative applications.

A) What is a cooperative application (1 point)?

B) Please give an example of a cooperative application that is also an active safety application (and explain why it is a cooperative active-safety application; 1 point). C) For the application that you choose, please give basic requirements for:

• Type of communication
• Transmission mode
• Frequency (update rate)
• Latency
• Data to be transmitted
• Range of communication

Answer 4

A) CSs (or C-ITS) are ITSs which rely on wireless communication to enable data exchange among
vehicles or between vehicles and the infrastructure. (Bishop 2005)

B) Of course, here many answers are possible. Cooperative curve speed warning: a road site
unit at a sharp curve exchanges information with vehicles approaching and passing the curve.
When the unit calculates that the speed of the vehicle is too high to pass the curve safety,
issues a warning for the driver to slow down. This implementation of CSW is a cooperative
application because it uses wireless communication to exchange info between the
infrastructure and the vehicle. Further, it is an active safety application because it senses a
critical situation (too high speed potentially leading to a lane departure) and it takes an
action (warning) to avoid for the critical situation to devolve into a crash.

C) For the application above:
a. Type of communication: point-to-point, two ways communication.
b. Transmission mode: periodic + on event
c. Frequency (updata rate): 10 Hz
d. Latency: 100-200 ms
e. Data to be transmitted: next page
f. Range of communication: 200 m
g. Data to be transmitted:

i. V2I: speed, position, acceleration, steering wheel position, mass, tires, etc…
ii. I2V: level of alert

Question 5

1) Please give a definition of naturalistic data (1 point)

Answer 5

Naturalistic data is data collected in the wild from road users attending to their daily routine.

Question 6

2) Compare naturalistic data to data collected from a driver simulator:
a. What are the main advantages of collecting data from driving simulators? (at least 2 examples; 1 point)

Answer 6

Safety and repeatability. We can expose different (or even the same subject) to the same critical situation and still accept the risk of crashing (because the crash would be virtual).

Question 7

2) Compare naturalistic data to data collected from a driver simulator:

b. What are the main advantages of collecting data in naturalistic driving studies? (at least 2 examples; 1 point)

Answer 7

Genuine behaviour and ecological validity. We capture the most authentic road user behaviour and we also expose our road users to the “full variability” of the traffic environment.

Question 8

3) Why do we care about naturalistic data, after all there are not many severe crashes in those datasets? (1 point)

Answer 8

We care about naturalistic data because, for active safety, it is more important to know what happened before a crash than afterwards. Naturalistic data may show how adaptation failures happen and whether active safety systems are successful in preventing adaptation failures.

Question 9

You are developing a new frontal collision warning (FCW) system.
1) What evaluation tool/s would you use to compare different human-machine-interfaces to deliver the warning? Why? (1 points)

Answer 9

In this case, a driving simulator is the most common tool because it makes possible to expose the driver to multiple and repeatable critical events without compromising safety

Question 9

c: Describe the billion miles problem for autonomous vehicles safety and provide two examples of how it can be addressed through design (2 points)

Answer 9

It is not possible to drive enough to show that a system is safe by observing the absence of accidents.

Math shows that an AD fleet of 100 test vehicles would have to drive hundreds of years to prove that the fatality rate is better than humans. Putting a system on the market (to increase testing volume) is not possible before it is shown safe.

Examples:
* Permissive driving – always know on what grounds an action is deemed safe.
* Balance run-time capabilities when making tactical decisions.
* Move uncertainties to the safe side of the outcome space.

Question 9

You are developing a new frontal collision warning (FCW) system

How could naturalistic data support the development of this new system, please give at least two examples. (Max 2 points; one per example)

Answer 9

3)
One example could be to use naturalistic data to verify, a posteriori, after introduction to the market, that the system work as intended. Another example would be to use the naturalistic data for counterfactual simulations as in the point above.

Question 9

You are developing a new frontal collision warning (FCW) system
2) What evaluation tool/s would you use to compare the safety benefit of your new FCW system with the older version? Why? (1 points)

Answer 9

In this case, a counterfactual simulation based on naturalistic data may be a good idea. In fact, this solution would enable a comparison in terms of avoided crashes (and reduced injuries) among the two systems

Question 10

This question is about the design of a (cooperative) active safety system (6 points)
Pick an active safety system of your choice and explain its function, in other words, a) please indicate
what is the safety critical situation that your system addresses (0.5 points) and what is the action that
it takes to avoid for the critical situation to devolve into a crash (0.5 points). b) Please indicate which
sensors, algorithms, and HMI/actuators your system uses (2 points). c) Please suggest how to make
this system cooperative (1 point). d) What would be the advantages (at least one example; 0.5
points) and the disadvantages (at least one example; 0.5 points) of this cooperative implementation
compared to the original one. e) Would this system still be useful for an automated vehicle able to
deliver level 3 automation (1)?

Answer 10

Of course, many answers are possible depending on the system of choice. Let’s assume a frontal collision wanning (FCW), a) FCW is a system that warns the driver in a car following scenario to avoid rear-end crashes. The critical situation may be indicated by the time to collision to the vehicle in front going below a set threshold and the action is a warning, typically a combination of acoustic and visual
cues.

b) A possible implementation of FCW would use a radar as a sensor and would base the threat assessment algorithm on the computation of time to collision. The decision-making algorithm may be
based on different thresholds on time to collision. Whenever time to collision would drop below a certain threshold, a warning could be issued; there may be different levels of warnings with their severity reflecting different thresholds. The HMI would normally include some sound, possibly played via the stereo system, a visual warning with dedicated lamps, and settings to tune the sensitivity of the warning. No steering or braking actuator is needed.

c) Replacing the radar with wireless
communication would make the system cooperative. (In other words, the two vehicles in the car following scenario exchange position information that is then used to compute time to collision.)

d)
An advantage of this cooperative implementation is the reduction of costs; a disadvantage is that the
communication requirements must be very high and the positioning extremely accurate for the system to operate correctly. With the current technology, the system would not work because the
combination of latency and uncertainties on the position would not allow a precise computation of time to collision.

e) FCW per se would be useful because in L3 the driver may still miss a critical situation and is still in charge when automation reaches its limit (of course, the cooperative implementation with the current technology may not be useful, independently of the automation level)

Question 11

Please, select an active safety system of your choice and answer the following questions:
1) What type of crashes does the system you selected address? (1 point)

2) Can you show a use case where your system would act in response to a critical situation? (1 point)

3) What is the architecture of your system? (Please, explain which sensors, algorithms, and humanmachine interface/actuators the system you selected employs. 2 points)

4) Can the system you selected be implemented in a cooperative way? If so, how? (Please, explain the communication strategy and which information is transferred by whom at what time. 2 points)

5) Will the system you selected be useful as automation increases? For each level of automation (from L1 to L5), please explain the extent to which the system you selected could still provide safety benefit to the
driver. (2 points)

Answer 11

Solution
Of course, the solution depends on the system you choose. I choose FCW.

1) Rear-end crashes.

2) In a lead-vehicle scenario, the vehicle in front brakes while the driver in the following vehicle is distracted. The FCW issues a warning, the driver looks up and brakes just in time to avoid the crash.

3) Sensor: camera + radar. Algorithms: threat assessment may be based on time to collision which is computed form range and range rate from the sensor fusion (radar+ camera). The decision making may be based on a threshold that keeps into consideration the driver response time. For instance, 1.8 s may be reasonable because it gives some time for the driver to redirect attention and ponder and execute an avoidance maneuver (most likely braking). HMI: the driver may be allowed to turn on and off the system as well as setting the threshold for the FCW. The FCW may use a combination of visual and acoustic warning to alert the driver.

4) In theory yes and, to be useful, positioning and communication may need to be better than they are today. Assuming that these technologies are sufficient, the two vehicles (lead and following) may share position (speed, and acceleration) and that may be sufficient to computer time to collision and replace the sensor info form radar and camera.

5) In L1 and L2, the system is expected to give the same benefit as in L0 (possibly more if the L1-2 would cause more opportunities for distraction). However, at least in L2, there would be some
longitudinal control going on which may prevent critical situations from ever happening. In L3, we have the same situation as L2 and it is hard to see a major advantage from FCW. From L4 to L5, FCW may be useless as the driver is not supposed to control the vehicle (and the vehicle may not
have controls).

Question 12

Ferruccio is driving a vehicle with Adaptive Cruise Control activated on a rural road and he is planning to overtake a pedestrian walking on the right side of the road. About 500 meters before reaching the pedestrian, Ferruccio receives a text on his mobile phone. Please answer the questions below:

a. Explain how adaptive cruise control operates (1 point)

b. Considering that Ferruccio is driving with Adaptive Cruise Control activated, which variable should be used to model the driver’s lateral control of the vehicle, at maneuvering level? Please use one of the models described in the lecture to motivate your answer. (1 point)

c. If Ferruccio takes the phone in his hand for about 3 seconds to answer the received text, how do
you expect that the perception of the pedestrian would be affected? Please use one of the models described in the lecture to motivate your answer. (1 point)

d. Assume that the vehicle is approaching the pedestrian and the Adaptive Cruise Control cannot
detect the pedestrian, leading to a possible rear-end crash scenario. Which variable can be used
as input to a driver model that aims to assess how long would it take for Ferruccio to react? Please
motivate your answer. (1 point)

Answer 12

Answers:
a. Adaptive cruise control (ACC) maintains a set speed (input by the driver) as long as no obstacle (such as a lead vehicle) are in front of the vehicle. When an obstacle appears on the forward roadway, ACC decreases the speed and, instead, keep a time gap to the obstacle constant. The
time gap is also input by the driver. In other words, ACC works as a cruise control as long as no vehicle is ahead and, when a vehicle is present, ACC keeps a constant distance or time gap to the
lead vehicle.

b. Based on the hierarchical control model, the maneuvering level focuses on anticipation (e.g. headway selection, lane position). The variable that should be used to model the driver’s lateral control of the vehicle at maneuvering level is the vehicle’s position in the lane (or the distance from the left/right lane).

c. Based on the zero-risk theory, the motivation to answer the text on the mobile phone would impair the perception of the pedestrian. This happens because the driver is shifting his gaze towards the phone, rather than looking at the road ahead (as a result of the motivation to answer
the text).

d. Looming can be used as input variable because looming has been shown to trigger driver’s reaction in rear-end crash scenarios.

Question 13

2) CAN bus (tot. 4 p). You are defining a CAN-bus frame that will include information about the
level of severity of a forward collision warning (high level = very severe warning).

a. If you want to include 5 levels of warning, how many bits will you need to reserve for
this information on your frame? (1 p)

b. You can decide between two IDs for this frame: 10101010101 and 00000111111. Which ID would you prefer? Why? (1 p)

c. Can you use the same frame to also include information about the current speed with a 0.5 km/h resolution? In other words, is the data field large enough to host both signals? (Please show some calculations to back up your answer; 2 p)

Answer 13

a) 3 bits

b) Here it does not matter which one you pick as long as it is clear that you know that the second frame will have higher priority than the first. (Of course, arguing for a lower priority when you want to convey a warning is hard, but a good argumentation would still give full score).

c) Speed for a car is meaningful from 0 up to 200, let’s call it 255 km/h. If you want a 0.5 km/h resolution, you need 512 levels which is 16 bits = 2 bytes. The data field of a CAN frame has 8 bytes, so you have plenty of “space” for this info on your frame. (Please notice that these are rough estimation and as long as your answer would be reasonable, other numbers, within the ball park of the ones above, may be reasonable)

Question 14

a. What is an active safety system? (1 point)

Answer 14

Active safety aims at preventing crashes and/or mitigating their consequences by recognizing safety-critical situations and taking action so that the situations don’t devolve into crashes.

Question 15

b. How does an active safety system differ from a passive safety system? (1 point)

Answer 15

Passive safety aims at reducing injuries from a collision by protecting the vehicle occupants.

Active safety aims at avoiding a collision or mitigating its outcome by taking action before the collision

Question 16

Please answer the following questions keeping in mind the SAE automation levels.

a. Explain what level 0 is (0.5 points).

Answer 16

Level 0 is manual driving, so no automation at all. (Some of you elaborated even more and provided a full SAE-level figure; that was appreciated but not of course required).

Question 16

Please answer the following questions keeping in mind the SAE automation levels.

b. Explain what level 2 is (0.5 points).

Answer 16

Level 2 assumes that both lateral and longitudinal control are automated within the DDT but not including OEDR.

Question 17

c. What is an intervention system? (0.5 points), and how does it differ from a warning system? (0.5 points)

Answer 17

Warning system: Inform the subject vehicle driver about the need to take action to avoid a specific hazard in the driving environment. [J3063].

Intervention system: Directly influence vehicle motions/actions for only the time needed to avoid/mitigate a specific hazard in the driving environment, and without necessarily requiring action by the vehicle driver. [J3063]

Question 18

Please answer the following questions keeping in mind the SAE automation levels.

c. Give an example of a commercial system providing level 1 automation (0.5), explain how this system works (sensor, algorithm, and human machine interface / actuators;1 point).

Answer 18

ACC (adaptive cruise control) deliver L1 automation by automating the longitudinal control within the DDT without performing the OEDR. ACC works as a cruise control when no vehicle is ahead. In vehicle following ACC maintain constant speed as cruise control as long as the vehicle ahead is faster, when the vehicle ahead is faster, ACC maintain a time or
space gap to the vehicle in front by adapting speed. ACC typically rely on a radar, the algorithm uses information about current speed and settings to activate brake and throttle so that speed and/or time gap are controlled. The human machine interface of ACC incudes, settings for speed, gap, and on/off. It also include visualization of current setting
on the dashboard. Finally, most ACC provide a warning when reaching max braking capacity

Question 19

Please answer the following questions keeping in mind the SAE automation levels.

d. Compare level 4 to level 3 and explain the main difference (1 point).

Answer 19

The main difference between L4 and L3 is who is responsible for the fall back; the driver in L3 and the machine in L4. This means that within a specific ODD, an L4 system can handle any situation (including failure and critical situations) without relying on the driver at all. The system alone, in L4, will always be able to reach a minimal risk condition.

Question 20

c. Naturalistic data includes objective and subjective data. Give 2 examples of objective data typically available in a naturalistic study (0.5 points) and 2 examples of subjective data typically available in a naturalistic study (0.5 points).

Answer 20

Objective: speed and acceleration

Subjective: weather and demographics

Question 20

Please answer the following questions keeping in mind the SAE automation levels.

e. Explain what the OEDR (object and event detection and response) is and how it helps the SAE classification of automation levels (1.5 points).

Answer 20

OEDR also includes driving events associated with system actions or outcomes, such as undiagnosed driving automation system errors or state changes. Up to L2, the feature is not responsible for the OEDR, from L3 and up, the feature performs the
OEDR.

Question 20

b. Why is naturalistic data so important for the development of active safety systems? (1 point)

Answer 20

Naturalistic data contains information about the driver behavior in the pre-crash phase, so it can help identifying the behavioral mechanisms leading to crashes

Naturalistic data can be used for counterfactual simulations of new active safety systems providing high ecological validity.

Naturalistic data can provide field evidence of the safety benefit from active safety.

Question 21

c. List three difference between data in crash databases and naturalistic data. (1point)

Answer 21

Crash databases include severe crashes whereas naturalistic data are often limited to minor crashes

Naturalistic data include the pre-crash phase, accident database may only have limited information about the pre-crash phase, either from interviews or from crash reconstruction.

Naturalistic data capture driver behavior (e.g. glance behavior, evasive
maneuvering, etc.), accident databases have very little information about the driver, often limited to demographics and medical records.

Question 22

a. Explain Level 3 from the SAE level of automation, more specifically (1 point)

i. Are both lateral and longitudinal control automated?

ii. Is the driver supposed to monitor the system during Level 3 automation?

iii. Who is responsible for potential fallbacks?

Answer 22

a. SAE level 3

i. Yes

ii. Yes

iii. The driver

Question 23

b. What is the operational design domain (ODD) and how does it relate to the differentiation of level 4 from level 5 automation? (1 point)

Answer 23

b. The ODD defined in which condition the system is supposed to provide automation. For level 4 the ODD is still limited, for level 5 the ODD is not limited.

Question 24

c. What are the main human-factor concerns for level 3 automation? (1 point)

Answer 24

c. The main issue with L3 is relaying on a human to take over from automation and perform fallback after not having being involved in the driving task for quite some time. This may not be safe because humans are not good at monitoring automation. The more a task is well performed by the automation the more error prone is a human:

Question 25

Cooperative applications make use of wireless communication to improve transportation.

b. What is DSRC and how does it relate to 802.11p? (1 point)

Answer 25

b) Dedicated short-range communication is a set of frequency reserved for cooperative applications, this is part of the physical layer enabling the 802.11p wireless data transmission.

Question 25

Cooperative applications make use of wireless communication to improve transportation.

a. Give an example of a cooperative safety application using V2V communication (1 point; please explain what is the purpose of the application and which data is transferred from where to where).

Answer 25

A possible example is emergency vehicle notification; in this case, an emergency vehicle, such as an ambulance, would broadcast its position, heading, and speed to the vehicles around. These vehicles should then warn the driver if they would be likely to end up on the ambulance path.

Question 26

Cooperative applications make use of wireless communication to improve transportation.

c. What is a Local Dynamic Map and how can it support cooperative applications? (please give at least one example in relation to one specific cooperative application;1 point)

Answer 26

c) An LDM is a database including information about the environment and other vehicles connected in a cooperative environment. Each vehicle has a unique LDM which is continuously updated as new information becomes available from other vehicles or the infrastructure. Cooperative applications relay on the LDM for threat assessment
and decision making.

Question 27

a. Provide a definition of “Driver model” (0.5 points)

Answer 27

Driver models describe driver’s behavior in terms of what he/she usually does and how he/she reacts in specific driving situation

Question 28

b. Explain the value of driver models for road safety research: why shall we use driver models? (1 point)

Answer 28

To understand driver behavior for different purposes:
Design/evaluation of active safety systems and autonomous driving:

o Predict safety benefits (e.g. decrease in crashes)

o Define settings(e.g. warning times)

o “Imitate” human behavior(autonomous driving)

Question 29

c. List the 4 main categories of driver models according to the classification provided during the lectures (0.5 points)

Answer 29

 Control models

 Hierarchical models

 Motivational models

 Information processing models