b'WHY AND HOW AUTOMATED VEHICLES CAN DRIVE LIKE PEOPLEBY HUEI PENG, DIRECTOR, MCITYLets summarize what and why AVs should drive likeBehavior competency testing to ensure automated Your first reaction to the title of this article may be, What? I thought theWITHOUT PROVENhumans and then focus on how. It should be clear byvehicles demonstrate they can handle a wide variety point was to design automated vehicles (AV) to reduce motor vehiclenow that we are not advocating for speeding, alcohol- of driving scenarios; andcrashes. What is the point if they drive just like human drivers? ROADMANSHIP,use, distracted driving, and road rage, but AVs must Corner cases, or test situations that push the limitsSELF-DRIVINGbe designed and programmed to create no hazard and of these highly advanced automated vehicles.Human drivers certainly are not perfect. In 2017, more than 37,133 peoplerespond well to hazards created by others. In addition were killed in motor vehicle crashes in the United States. [1] More thanCARS MAY NEVERto being safe, being normal is important to sustainThe primary goal of the Mcity ABC Test is to assess 10,000 were due to alcohol-impaired-driving, and more than 9,700 werean efficient mobility system. the performance of HAVs, including both safety and speed-related, despite years of police enforcement.EARN CONSUMERroadmanship. HAVs should display behaviors similar TRUST ANDSohowdoweprogramroadmanshipintoAVs?to those of human drivers without being too aggressive Still, even with these examples and other forms of driver error, humanAndhowdowemeasureitasacomplementaryor conservative in comparison with human drivers.beings are remarkably safe and adaptive drivers. There are 1.16 fatalCONFIDENCEperformance metric to safety? Mcity, a public-private crashes every 100 million miles. These miles vary from simple lane-keepingR&D partnership led by the University of Michigan,ThebehaviorcompetenceportionoftheABCTest and car following to complex interactions at intersections, ramp entrances,AT A LEVELis tackling this challenge. Mcity is funding researchprotocol is focused on executing 16 driving scenarios.roundabouts, and parking lots, often involving multiple road users. HumanNECESSARY TOprojectstohelpdefineroadmanshipmeasuresforIn many of these driving scenarios, a challenge vehicle drivers can generally handle road work, closures, potholes, parked cars,highly automated vehicles (HAVs).willbeprogrammedtofollowaspecificpathata and jaywalkers easily, and under certain situations, they have no problemSUPPORT THEIRprescribed speed until it reaches a conflict point with breaking the law, such as crossing a double-yellow line to avoid parkedHAVs include Level 4 and Level 5 AVs on a six-level scalean HAV. The scenarios include behavior such as:delivery trucks. Arguably, not all of these adaptive behaviors should beWIDESPREADdefined by SAE and adopted by the U.S. Department programmed into AVs, but some should. ADOPTION. of Transportation. Level 4 vehicles dont require human Performing a low-speed merge; controlwhenoperatedwithindefinedareas;level5 Following a lead vehicle, including in stop-and-goTodaysAVsaregenerallyprogrammedtobesafeunderadefineddriverless vehicles can go anywhere, anytime. situations;operational design domain (ODD). While safety is and should remain the Detecting and responding to lane changes;top priority, safety alone is necessary but not sufficient for an AV. ConsiderMcity researchers use naturalistic driving data they Making appropriate right of way decisions at unprotected left-turns and entering a roundabout as examples. An AVcollect to study how human drivers react in various traffic crosswalks;that is safe but fails to take many safe gaps is unacceptable. Similarly, ascenariosasameanstolearnhowHAVsshouldbe Detecting and responding to pedestrians in thevehicle that brakes unexpectedly and more harshly than typical human- programmed to respond in similar situations. road, not at a crosswalk; anddriven vehicles, especially when there is no clear reason to do so, can be Keeping a safe distance from pedestrians anda nuisance or hazard to other road users or to onboard passengers. InMcity is developing a test-track based procedure, the bicyclists on the side of the road.both situations, being safe is not reason enough for not exhibitingMcity ABC Test, that could emerge as the basis for a good roadmanship. voluntarystandardforsafetytesting.Thethree-partOther Mcity-funded research projects will go a step procedure includes: furtherandattempttoanswerthequestionsraised In a report about AV safety published in 2018, RAND Corp. [2] definedby the behavior competence test data: Is the HAVs the term roadmanship as the ability to drive on the road safely without Accelerated evaluation covering the driving reaction safe? Is it safe enough? Is it too safe and not creatinghazards,andrespondingwell(regardlessoflegality)tothe scenarios responsible for the most common motor demonstrating proper roadmanship? The answer lies hazards created by others. An AV constantly slamming on its brakes to vehicle crashes; in how the HAV scores against the range of human avoid hitting something, for example, would be at best annoying and atresponses: is the HAVs reaction what another driver worst a hazard. 19'