Please enable JavaScript.
Coggle requires JavaScript to display documents.
Task 8 Human Factors (Wilson & Russell (2007) Unmanned Air Vehicle…
Task 8 Human Factors
Wilson & Russell (2007) Unmanned Air Vehicle (UAV)
Variables
OFS
--> operator functional state
--> threshold determined by increasing UAV speed in either easy or difficult radar image
ANN
--> artificial neural network
SWR
--> successful weapon release
UAV
--> Unmanned arial vehicle
research question
ANN Aid based on psychophysiological measures improve performance?
Individual OFS threshold = better performance than group OFS threshold?
ending of ANN aid based on psychophysiological measures = better than ending of AAN aid based on end of subtask?
Methods
10 participants
--> 5 high performing 5 low performing
radar image
--> had to mark target houses before UAV arrived + releasesd weapons
easy (less terrain + houses)
hard (lots of terrain + lots houses)
Speed of UAV
--> increased until only able to mark 25 to 30% of required targets before UAV arrived
this became the person's threshold of OFS !!
averaging eachs person OFS threshold (so speed limit of UAV for functioning) together = group OFS
Health of UAV
--> communicated via text messages
ANN aid
--> reducing UVA speed by 50% (more time to target houses)
--> unifying UAV health messages in one window (easier to monitor)
Results:
Whole group
Highest improvement but no significant difference between:
Individual aid (50% increase reported)
group aid (35% increase reported
random group aid
leave on aid (until subtask finish) (35 % increase reported)
low performance group
individual aid + leave on aid = highest improvement + same significance
surprisingly leave on aid = significantly better than group aid!!
IMPORTANT:
--> group aid less usefull cause high performance operators pull the average speed of drones above the cognitive performance level of waht low performing operators can handle !!
--> so they are already completely cognitively overloaded but the ANN aid will only set in much much later thus it wont help them much because even with the aid it UAVS still too fast
high performance group
group aid = highest significance +biggest improvement of 95%
individual aid + leave on aid = same effect and second highest significance
IMPORTANT:
--> group aid = more useful because low performing operators pull the average speed of the drones down so the cognitive performance load is way below what a high performance operator can handle so they are not cognitively fully loaded.
--> so they are cognitively underloaded and on top of that the AAN aid engages much earlier than they would need it thus they perform super easy breezy on this one because for them the UAV seems slow so they have all the time in the world :3!!
limitations
NO BONFERRONI CORRECTION IN ANOVA TEST!!
--> so some significant results might not be significant after all !!
--> biggest flaw cause no bonferroni correction can double chances for significance or even show something that isn't even significant as significant!! (fmri and atlantic salmon study !!!)
Small sample size !!
Leave on condition has exactly the same effect as individual aid
leave on condition in whole group = as significant as individual aid
ANOVA not performed on any possible combination of experiment conditions !! only the one ones the experimenters wanted to highlight so their results and conclusions might be completely wrong!
Human interaction with automation
What is automation?
device or system that partially or fully carries out operation that could previously only be done by humans
Levels of automation
continuous from:
---> low (fully manual/human interaction)
---> to high (fully autonomous/automatic)
adaptive
--> level of automation can vary with changing demands
--> e.g. high if X but low if Y
#
4 classes of function
--> each of these can have a
level of automation ranging
form low to high on a continuum
--> levels can be adapted based on demands
#
modeled after human functioning (sequential)
sensory processing to -->
perception/working memory to -->
decision making to -->
response selection + execution
1. information acquisition
#
-->registering input data
--> analogous to sensory processing
High level = filtering data already
Low level = just registering
2. information analysis
--> analogous to perception/ working memory
#
low level = extrapolation or prediction of data over time
high level = combining several pieces of information into one (integration)
3. decision + action selection
--> choosing between different alternative options
--> ACT-R + production rules = great example of this!!
#
low level = suggest decision
high = make decision without human interaction
--> usefull in high stake environments (nuclear reactor or tesla car accident prevention))
4. action implementation
--> execute the selected decision/choice/action :3 !!
#
low level = manual user input required
high level = executes action without any human interaction
--> usefull in high stake environments (nuclear reactors or tesla car accident prevention)
Model evaluation Criteria
Primary
human performance
Mental workload
--> well designed automation reduces cognitive workload and operators perform better
--> badly designed automation increases cognitive workload and operators perform worse
situation awareness
--> too much decision making automation may reduce awareness of dynamics of the system / how it works or everything he would need to manually if assistance breaks
Complacency
--> if automation highly but not perfectly reliable, human operator may blindly trust and stop monitoring machine thus potentially missing critical errors machine makes:3 !!
Skill degradation
--> if machine always makes all decisions humans becomes less good at doing it by themselves over time!
secondary
Cost of decision/action
--> based on risk
risk = cost of error * probability of error
-
->
if low
= full/ high automation warranted
ALWAYS warranted if:
!!
--> if highly time critical and important decisions = full/high automation (e.g. nuclear power plant meltdown, tesla car preventing crash :3!!)
reliability of automation
--> influences human trust!
too high
= people stop paying attention to it because they expect it to work flawless --> might lead to them overseeing a critical error!
too low
= people don't trust in the machine which negates all positive influences the automation process could have had!
Model suggestion
Information aquisition
High level of automation IF reliable!!
Information analysis
High level of automation IF reliable!!
decision + action selection
low-ish for high risk functions
--> exception below
#
High for low risk function
ALWAYS HIGH FOR
--> if highly critical and time constraint
e.g. nuclear power plant meltdown, tesla car crash prevention
action implementation
medium level for high risk functions + high level/ important decision
--> exception below
#
ALWAYS HIGH FOR
--> if highly critical and time constraint
e.g. nuclear power plant meltdown, tesla car crash prevention
Presentation Ereaders
Smartphones
Baking automation