Nothing Special   »   [go: up one dir, main page]

0% found this document useful (0 votes)
12 views16 pages

L104 RemovingHumanDriver

Download as pdf or txt
Download as pdf or txt
Download as pdf or txt
You are on page 1/ 16

Implications of

Removing the
Human Driver
https://goo.gl/YUC5oU

© 2021 Philip Koopman 34


See: https://users.ece.cmu.edu/~koopman/toyota/index.html © 2021 Philip Koopman 35
https://bit.ly/2m9cQRm

Huang fatality;
crash into
concrete median.
© 2021 Philip Koopman 36
© 2021 Philip Koopman 37
Can Humans Safely Supervise Autonomy?

https://goo.gl/ZFCYzD

https://goo.gl/VTFW9d

https://goo.gl/kgRq71
© 2021 Philip Koopman 38
The 94% Human Error False Narrative
 Where did “94%” come from? “94%”
 “The critical reason was
assigned to drivers in an estimated
2,046,000 crashes that comprise
94 percent of the NMVCCS crashes
at the national level.
However, in none of these cases was
the assignment intended to blame
the driver for causing the crash.”
[DOT HS 812 115]
 Looking a little deeper:
https://www.nhtsa.gov/technology-

 74% of driver errors were “recognition” or “decision” errors


innovation/automated-vehicles-safety

 And software driver must handle the 6% of no-driver-involvement crash causes


– Tires, brakes, drivetrain failures
© 2021 Philip Koopman 39
Humans Are Amazing Fault Mitigators
 Other side of the “94%” coin – people prevent crashes too

 Toyota uncommanded acceleration – most saved by human


 89 deaths, 57 injuries as of May 2010
 6,200+ NHTSA complaints [https://www.cbsnews.com/news/toyota-unintended-acceleration-has-killed-89/]

 GM brake issues – most saved by human


 293 injuries, 2111 crashes
 10,861 NHTSA complaints https://www.nytimes.com/1999/07/22/us/gm-admits-brake-flaws-after-inquiry.html

 Will an ADS be as successful at fault mitigation as humans?


 ADS will need to deal with heavy-tail issues © 2021 Philip Koopman 40
Automotive Software Quality Issues
Software
defects &

}
software
integration
problems

Source: Stout 2020 Automotive Defect & Recall Report


IEC is integrated electronic components (hardware) © 2021 Philip Koopman 41
Automotive Software Has Defects
 Small sampling of NHTSA recalls (i.e., confirmed bugs)
 See: https://betterembsw.blogspot.com/p/potentially-deadly-automotive-software.html
 21V-071 Vehicle unexpected pulls to one side during evasive maneuver
 20V-213 Remote smart park continued motion after failsafe activation
 19E-070 Anti-rollback software causes unexpected vehicle motion
 19V-539 Forward collision avoidance does not detect stationary vehicle
 19V-351 Regenerative braking failure reduces deceleration
 19V-075 Transmission unexpected downshift to first gear causes loss of control
 18V-621 Automatic braking cancelled / ABS locks up wheels
 18V-607 Active Lane Keeping Assist does not intervene in lane departure
 17V-713: Engine does not reduce power due to ESP software defect
 17V-686 and MANY others: Airbags disabled
 15V-460 and others: Airbags deploy when they should not
42
© 2021 Philip Koopman
Example Required ADS Fault Handling
 Tire blowout/wheel detachment
 ADS: perform controlled stop (or run-flat tire operations)

 Service brake failure


 ADS: downshift/regen braking, apply parking brake, runaway ramp

 Catastrophic sensor failure


 ADS: dead reckon to stop using most recent object trajectories

 Uncommanded acceleration
 ADS: de-energize engine/motors, apply forceful brakes

 Main battery fire


 ADS: shed electrical load, stop vehicle, passenger evacuation
© 2021 Philip Koopman 43
Controllability Without A Human Driver
 What happens when there is no

??
human to exert controllability?
 Own vehicle human driver?
 Other vehicle human driver?

 Some combination of:


 ADS will need to control faults to
attain C1 or C2
 Vehicle will have to upgrade
subsystems to C3 (“uncontrollable”)  ISO 26262 Driver Controllability:
 C1 = Simply controllable
 Potential for significant ASIL
 C2 = Normally controllable
increase across whole vehicle
 C3 = Difficult / uncontrollable
 Many ADS control requirements
© 2021 Philip Koopman 44
No Human Driver to Blame
 “Computers won’t drive drunk” .. but …
 Drunk/DUI is only 28% of fatalities (US 2019)
[https://crashstats.nhtsa.dot.gov/Api/Public/Publication/813060]

 Automated Driving Systems (ADS) will


likely make different mistakes
– Perception/classification errors
– Brittle in face of surprises (unknown unknowns)

 What happens with ADS “driver error”?


 Every AV crash is a product liability lawsuit
waiting to happen
 Eventually, no human driver to absorb blame
– What about Driver monitor system (DMS) failures?
© 2021 Philip Koopman 45
© 2021 Philip Koopman 46
Operations & Human Interactions
 Drivers do more than just drive
 Occupant behavior, passenger safety
 Detecting and managing equipment faults

 Operational limitations & situations


 System exits Operational Design Domain https://bit.ly/2GvDkUN

 Vehicle fire or catastrophic failure


 Post-crash response

 Interacting with non-drivers


 Pedestrians, passengers
 Police, emergency responders https://bit.ly/2PhzilT

© 2021 Philip Koopman 47


Lifecycle Issues
 Handling updates https://bit.ly/2IKlZJ9

 Fully recertify after


every weekly update?
 Security in general

 Vehicle maintenance
 Pre-flight checks, cleaning
 Corrective maintenance

 Supply chain issues


 Quality fade https://bit.ly/2VavsjM

 Supply chain faults Is windshield cleaning fluid life critical?


© 2021 Philip Koopman 48
Changing Role of
Human Driver
 No human driver to blame for crashes
 ADS handles vehicle equipment failures
 ADS handles non-ADS software failures

© 2021 Philip Koopman 49

You might also like