Discussion Guide For Automated And Connected Vehicles, Pedestrians, And .

What technologies are being developed to help AVs detect pedestrians and bicyclists?Below is a brief summary of some currentsystems and their pros and cons:Machine vision systems are relativelyinexpensive to implement (as cameras areless expensive than lidar-based technologies),but are challenged by rapid detection ascurrent systems are still significantly slowerthan human perception. They are alsounable to compensate for obstructions andare susceptible to conditions that can impaircamera performance, including darkness andadverse weather (fog, rain, snow, etc.) aswell as camera lens degradation.Lidar allows fine-detail mapping and avoidssome of the environmental issues faced byand provide a larger safety margin. This willprovide a service both for human drivers in thepresent as well as the more automated machinedrivers of the future. Some of the most effectivesafety treatments, such as separated bike lanes,improvements in lighting, pedestrian crossingislands, and gateway treatments (FHWA, n.d.),will likely aid in making it easier to detect orpredict the presence of pedestrians and bicycles.Additionally, wireless beacons could theoreticallyaid in detection as well as connect to infrastructure(P2I) to affect signal timing and prioritization forpedestrians and bicyclists. However, considerationmust be given to people who may not be carryinga device by choice or because they do not have themeans to own a device (see #2: the V2X problem).Current and needed research: Multipleapproaches are currently under development toimprove automated systems’ ability to detectand identify pedestrians and bicyclists, which willenable better warning and avoidance technologies(see sidebar). These approaches include onboardsystems such as camera-based machine visionsystems (Harris, 2015; Hsu, 2016), radar (Siemens,n.d.), and advanced lidar (Ross, 2017; Navarro,6machine vision (for example, it is unaffectedby low light), but current lidar hardware isbulky and expensive (although upcomingsystems have the potential to ameliorate thislimitation) and it is vulnerable to problemsin weather conditions involving rain, fog,snow, and dust.V2X beacons are based on variouscommunication technologies that wirelesslyconnect pedestrians or bicyclists and vehicles.They present a way to positively identifypedestrians or bicyclists no matter what thelight, weather, or obstructions, as well asinfer their trajectories. However, there areconcerns surrounding this technology as well(described more in #2: The V2X Problem).et at., 2016), as well as networked solutionsincluding wireless V2P pedestrian identificationbeacons (USDOT, n.d.-b.; Volpe, 2017). Avideo analytics project, led by the Instituteof Transportation Engineers (ITE), is currentlyunderway to facilitate improved machine learningbased on traffic camera footage. Future ADSapplications could build upon the data and what islearned through this project. Additional researchis needed to evaluate the pedestrian and bicycledetection capabilities of different ADS sensorsystems under various conditions, including: lowlight, glare, adverse weather conditions, visuallycluttered landscapes, crowded streets, amidhorizontal and vertical curves, and obstacles suchas parked cars. It will be important to evaluatethe technology in terms of its ability to detectand predict people with diverse geometric shapes(including people in wheelchairs, different types ofbicycles, etc.). Researchers also need to improveupon methods to infer pedestrian or bicycle travelbehavior and directional intent. Studies to modelor predict both crowd and individual behaviorbased on observable characteristics (such as bodyposture, travel speed, and direction) could beuseful in ADS development.

Figure 3. Vehicle to pedestrian or bicyclist technologies (V2X) could aid in detection of and communication withpedestrians and bicyclists, but an array of technology and equity issues must still be addressed.#2: The V2X ProblemWhat it is: Connected technology such asvehicle-to-vehicle (V2V), vehicle-to-pedestrian(V2P), and other V2X technologies represent thepotential for safety improvements utilizing shortrange communication to inform roadway usersand the infrastructure itself about the presenceand status of road users such as pedestrians.V2X technology, however, suffers from multipleproblems that currently limit its use in improvingdetection of and communication with pedestriansand bicyclists. These issues include poor locationaccuracy (especially in urban canyons), inability toforecast maneuver intentions, serious challengesin minimizing false positives and false negatives,privacy issues, and issues surrounding driver/vehicle expectations when confronted with peoplenot carrying an active beacon due to cost, choice,or system failure. Two key challenges will be in (1)designing systems that can be beneficial even ifthe connected technology is not ubiquitous; and(2) avoiding systems that create new blind spotsand unintended consequences (e.g., by presumingthat all road users are connected or by prioritizingconnected signals over direct perception).Why it matters: Multiple threat situations, wherea driver passes another vehicle that is blocking hisor her view of a crossing pedestrian, often leadto serious pedestrian crashes. Many other crashtypes result when sight distances are limited orpedestrians or bicyclists are obscured from viewby objects such as trees, utility poles, or parkedvehicles. Given the limitations of current detectiontechnologies (see #1: The Detection Problem),connected technologies (using cell phones orperhaps built into or mounted on bicycles) couldimprove driver or vehicle detection of pedestriansor bicyclists in important situations, irrespectiveof the level of automation. V2P beacons could7

also play a significant role in improving safetyfor particularly vulnerable pedestrians, includingpeople who may act unpredictably (such aschildren and people with mental illness) as wellas people who need additional crossing time atsignalized intersections (such as children, seniors,and people with disabilities). Wireless beaconscould theoretically connect to infrastructure(P2I) to affect signal timing and prioritization forpedestrians and bicyclists. This technology couldhelp compensate for the narrow detection profileof bicyclists, as well as provide warning when abicyclist is approaching an intersection and maynot be able to stop in time.Policy implications: All people have a right totravel on public streets safely, so ultimately ADSand connected systems must find a way to detectand respond to all road users, not just thosecarrying devices. Consideration must be given topeople who may not be carrying a device by choiceor because they do not have the means to own adevice. For example, children may be less likelyto have smartphone-based V2P systems to warnof their presence, resulting in a case where V2Xtechnology alone will not likely produce safetybenefits for child pedestrians. It is also critical thatequity considerations factor into the discussion,such that differential safety benefits will notaccrue for pedestrians and bicyclists with themeans to afford advanced wireless communicationsystems relative to those who cannot.8Current and needed research: The IntelligentTransportation Systems Joint Program Office (ITSJPO), the Federal Highway Administration (FHWA),and NHTSA has been conducting V2P researchand investigating a range of applications. For V2Xapplications, there is a need to better understandthe limitations of their application (such as cost,limited accuracy of positioning information, noability to predict intentions, privacy concerns,trust, expectation, use, reliability, and overreliance) as well as their potential benefits (suchas improved detection or crash prevention).There is also a need to evaluate the availabilityand practicality of V2X technologies for all typesof pedestrians and bicyclists to ensure that thebenefits of these technologies are equitablydistributed (i.e., to people of all ages, incomes,educational levels, physical abilities, etc.).

Figure 4. How pedestrians and bicyclists will learn to identify and communicate effectively with ADS is an open question.#3: The CommunicationProblemWhat it is: Communicating intentions betweenroad users occur in a variety of ways, fromfacial expressions to head movements to handgestures. In an ADS future, this culturally-based,human-centric communication will graduallybe supplemented and eventually replaced by ahuman-machine interface (HMI) that may be moreor less transparent and intuitive. Consider, forexample, a case where a driver wishing to turnacross a sidewalk to enter a driveway or parkinglot encounters a pedestrian whose trajectory islikely to intersect the vehicle’s. In this case, thepedestrian likely has legal right-of-way, but toensure her safe travel she is likely to confirmthat she has been seen by the driver. Theircommunication would involve a complex andculturally-guided series of interactions, includingfacial expressions (e.g., smiles, raised eyebrows,etc.), and gestures (e.g., a horizontal wavemeaning “go ahead” or a vertical wave meaning“thanks”). Many communication cues used todaycould be absent or presented differently in ADS.This shifts the communication from interpersonalto human-machine, with humans on both sidesof the interaction being affected. In the future,pedestrians and other road users will have tofigure out when a vehicle is being controlled byADS, how to establish trust that they have beendetected, and how to communicate regardingexpected behaviors, and when to proceed. Thismay be particularly challenging for people withdisabilities, especially vision-related disabilities.Automated vehicles may also require significant“training” or programming in order to understandhuman communications and predict behaviors(such as knowing how to interpret a hand gesturewhen a bicyclist signals intent to turn or stop,or how to interpret from body language when9

a pedestrian standing on the side of the road istrying to cross the street versus simply waitingon a bus).Why it matters: Interactions between AVs andbicyclists and pedestrians (and all other roadusers) need to be transparent, socially acceptable,and efficient. Consistent communication standardsare needed to ensure both acceptance and safety,particularly across cultures where right-of-waylaws and norms may vary considerably (see #4:The Right-of-Way Problem). ADS vehicles will needto communicate intent with symbols, text, and/or sounds, and take into account the needs ofpeople with hearing or vision impairments. Thesesymbols can’t be inconsistent with the Manual onUniform Traffic Control Devices (MUTCD), whichregulates roadway signs, signals, and pavementmarkings. For example, it would not be acceptableto have round, green stop signs mounted on thefront of AVs to indicate the need to stop. Further,communications must be consistent across all ADSvehicle types to prevent confusion and risk.Policy implications: Communication issues arelikely to be made more challenging by mixed fleetswith many different HMIs and styles of operation.Data and “blueprint” sharing may be necessary(though not likely to occur voluntarily) to ensurethat communication systems are consistentlyintegrated and tested across all makes and modelsof vehicles, and understood by vehicle users andpedestrians and bicyclists alike.10Current and needed research: There is ongoingU.S. Department of Transportation (USDOT)research surrounding automated vehiclecommunication of intent with shared road users.Additional research is needed to explore howpeople desire to communicate and will actuallycommunicate with AVs, and how this will affecttheir behaviors and interactions in the future.It is also critical that cross-cultural factors beconsidered, as rules, customs, social norms,languages, and traffic laws will vary not onlybetween countries but within diverse countriessuch as the United States. It is an open questionhow a car developed and tested by a manufacturerin one country will be regionalized for behaviorand communication styles that make sense topedestrians and bicyclists in specific markets, letalone regions where the vehicle may travel overits lifespan. Additionally, while numerous studieshave sought to understand what the driving publicwants out of future vehicles, there is currently alack of research on the needs, desires, and comfortlevel of pedestrians and bicyclists in relation totraveling around and communicating with AVs.See #5: Passing Problem for one recent survey of abicycle advocacy group’s membership.

Figure 5. There are many unanswered questions with respect to how ADS will adhere to varying laws requiringyielding right-of-way to pedestrians, and the ripple effects that ADS interactions will have on human driver andpedestrian expectations and behaviors.#4: The Right-Of-WayProblemWhat it is: This issue is closely related to #3: TheCommunication Problem. Social customs andcommunications that govern giving right-of-wayto pedestrians vary from place to place. State andlocal laws also vary with respect to who has theright-of-way in crosswalks in different settings,and even in the definition of a crosswalk (e.g.,marked versus unmarked crossing) and wherethe pedestrian must be with respect to it (e.g.,with a foot in the crosswalk or approaching it).Driver failure to give right-of-way to pedestriansat legal crossings is a leading cause of pedestriancrashes (Schneider & Sanders, 2015). Currently, itis not well established how AVs will yield right-ofway to pedestrians or what other communicationchallenges, behavioral adaptations (from otherpedestrians or human drivers), or other unintendedconsequences will arise as more of these AVpedestrian interactions take place. For example, ifan ADS is designed to always stop for pedestrians,there is the potential for pedestrian behavioraladaptation (e.g., “gaming” the limitations of thesystem) or over-trust of the technology, whichcould increase the risk of conflicts and crashesin mixed fleets (i.e., when ADS vehicles yield atcrosswalks and human drivers often do not). Itis also possible that zones with high levels ofpedestrian activity and many pedestrian crossingswill make automated vehicle movement inefficientand impractical.Why it matters: Theoretically, having more ADSon the roadway that routinely yield to pedestriansmay lead human-driven vehicles to follow suit,further establishing yielding to pedestrians asa social norm in more places and leading to11

safety benefits. Combined with advancements inconnectivity, it is conceivable that pedestrians maybe able to cross at locations and/or times that aremore flexible and convenient than going to thenearest marked crosswalk and waiting for asignal, thus enhancing mobility as well. Thesescenarios raise implications for how transportationagencies can approach multimodal trafficmanagement in order to facilitate safe andefficient interactions between all road users andsupport social and economic vitality in a varietyof development contexts.Policy implications: Automobiles, regardlessof the level of automation, should give rightof-way to pedestrians at legal crosswalks ascurrent laws require. If right-of-way laws differacross jurisdictions in which an AV operates, thenadditional challenges may arise. Even when apedestrian doesn’t have the legal right-of-way,vehicles should make every effort to avoid a crashwith a pedestrian. Regulation of the industryto create standardized “rules” can also createconsistency, which is vital for interaction withother road users and may have ripple effectson liability in the event of a crash. Having avariety of vehicle yielding behaviors across trafficsituations could cause significant confusion orrisk in a mixed-vehicle fleet if a proportion ofvehicles are controlled by fixed programmingfrom a manufacturer while the rest of the drivingpopulation is guided by social norms that mayconflict with formal regulations. Either ADSprogramming will need to have some amount ofbuilt in learning and flexibility, or localized socialnorms will have to undergo some periodof adjustment to a more common set ofstandards, or both.12Current and needed research: Due to the rapidpace of technology development under proprietaryconditions, there is currently little information thatis publicly available on how auto manufacturersand software developers are creating thealgorithms used to govern ADS vehicle behavior,and what safety implications these may have.According to nuTonomy’s Chief Operating OfficerDoug Parker, “Essentially, we establish a hierarchyof rules and break the least important,” he said(Bhuiyan, 2016). It will be important for thepedestrian, bicycle, and safety community tohave a voice in determining the values that drivethe “hierarchy” of AV navigational rule-making,in particular the rules used to govern when AVsgive right-of-way to other road users. Research isneeded to independently test how ADS maneuverand give right-of-way to pedestrians and bicyclistsand whether changes to state or local statutemay be needed to enhance safety and consistencyduring these interactions.

Figure 6. The ways in which ADS are programmed to pass slower-moving cyclists, within the confines of existingtraffic laws, will be an important issue to address.#5: The Passing ProblemWhat it is: Recent studies have shown thatmore than one-third of bicyclist fatalities involveimproper overtaking maneuvers by the driver(Schneider & Stefanich, 2016; McLeod & Murphy,2014). Drivers passing bicyclists too closely is acommon and unpleasant experience for manyriders, which has been shown to reduce interestin bicycling (Sanders, 2015). There is currentlysignificant variation among state laws with respectto the minimum passing distance required aroundbicyclists. Little to no research exists to show whatpassing distance is most safe or comfortable forthe bicyclist at different travel speeds, or howthese laws have been enforced or affect actualpassing distances. In addition to variations inpassing laws, there is even more variation in thepresence and quality of existing on-road bicycleinfrastructure (e.g., wide shoulders, bike lanes,and shared lane facilities), which determinesthe level of separation between bicyclists andmotor vehicles.Why it matters: ADS will need to be equippedwith advanced detection, prediction, andavoidance maneuvers to safely navigate aroundbicyclists, whose behaviors may include weavingbetween vehicles, legally riding in the middle ofa travel lane at a slower speed than other traffic,and even cases where a bicyclist has crashed.Additionally, ADS will need to have a set o

In this paper, we use the term "automated driving systems" (ADS) to refer to vehicles with SAE Level 3 automation or higher. We use the term "automated vehicle technologies" (AV) when referring to automated vehicles in general. The following terms and technologies are referenced throughout this paper: Automated Driving Systems (ADS) are