Communication between automated vehicles and vulnerable road users in future traffic
Summary by:
Marilee Nugent
University of Florence | UNIFI · Dipartimento di Ingegneria Industriale
In the new opinion paper ‘Vulnerable road users and the coming wave of automated vehicles: Expert perspectives‘ 16 experts in human factors and vehicle automation technology weigh in on the hot issues and challenges of integrating vehicle automation and connected technologies in urban traffic. How will AVs and vulnerable road-users - pedestrians, cyclists, motorcyclists - interact? What messages need to be communicated between them and how?
Can automated vehicles (AVs) make urban traffic environments safer for vulnerable road users (VRUs)? By reducing human error - and bad decisions - the hope is that the majority of causal factors in vehicle-to-VRU crashes can be drastically reduced. A vehicle programmed to obey road rules would not run amber or red lights, suddenly try to make a turn from the wrong lane, exceed speed limits in pedestrian dense areas, fail to stop at an occupied zebra crossing, or drive under the influence of alcohol.
Sounds promising on paper, but what are the practical challenges? Most of the automation research has been done on highway environments where, although traffic flows at high speeds, road user interactions are more predictable than in urban environments. In cities and towns, however, the modes of travel are more varied and the interactions are much more complex. Through both instinct and cultural training, we exploit the exchange of many different forms of signals and social cues to predict what others driving, walking, cycling, running intend to do next, to help us make decisions about our own movements. So too, AVs will have not only to detect VRUs but predict their behaviour and communicate the vehicle’s intentions.
In this new opinion paper, 16 researchers in human factors were interviewed for their expert personal opinions on the current status of technological developments, bringing us up to date on the ongoing challenges to be addressed for safe interaction between VRUs and AVs in the future urban traffic.
How automated will future traffic be? The evolution of mixed automated traffic
The consensus opinion was that it will be decades before we see a significant presence of SAE Level 4 vehicles in the traffic mix. The introduction of Level 5 fully automated vehicles will have a much longer timeline. Thus, we may only see Level 5 AVs in restricted use settings as is currently seen in some industries, or as very slow-moving passenger shuttles in segregated environments. Some experts were of the opinion that full driving autonomy will never be achieved given the difficulty in achieving adaptability for all weather, road and traffic participation scenarios.
Some experts pointed out ambiguities in the definitions of the automation levels creates some conceptual fuzziness. For example, a Level 4 public transport vehicle might be autonomous for specific routes, whereas a personal AV may be operating at Level 4 in certain contexts, but requires human takeover and thus reverts to Level 3 control in others. Several researchers expressed concerns about the safety of Level 3 vehicles because the division of monitoring and control tasks between the driver and the system, respectively, has been found to confuse drivers as to their role which has led to misunderstanding, misuse of, or overconfidence in automated systems. This division also causes drivers to become bored and inattentive, compromising takeover reactions in emergencies. These problems have led to crashes. Some experts therefore suggested that this step should be skipped, or that for now, it would be better to limit AVs to freeways.
Shared control is put forth as an intermediate step towards full automation, but this term should be clarified as usage has been inconsistent and should not be confused with Monitoring in which the occupant is expected to be attentive to the vehicle and environment for hours but do nothing to control the vehicle. Trading control has its own downside as when switching from a non-monitoring state to retake control in response to an emergency takes much longer than the reaction of a conventional driver. Instead, true shared control may be a better step in the evolution to full automation, described as the horse and rider model: the rider controls the goals and pace but the horse takes care of how the low-level control is specified. The rider can intervene to make the horse do things it cannot on its own, and the horse can take over if the rider falls asleep or is otherwise occupied or incapacitated.
Challenges facing AV presence in pedestrian-dense areas
As AVs will have to communicate their intentions to other traffic participants, they will also have to be responsive to VRU behaviour. How can precedence be apportioned to ensure VRU safety while maintaining traffic flow? The level of conservatism for giving precedence to VRUs is yet to be established. Changes in VRU behaviour as we interact more with AVs could pose new challenges and problems. For example, if we learn that AVs will always stop and give precedence to pedestrians, indiscriminate jay-walking could seriously affect the traffic flow. Developers must still refine the technology that enables AVs to “see” around other vehicles and obstacles for VRU detection, as well as the algorithms that can predict their future movements for crash avoidance. The best solution in some high-density pedestrian areas may be to ban AVs altogether.
AV-VRU communication & connectivity
As more of the driving functions are taken over by the vehicle and the driver/occupant is freed up to do other tasks, traditional human-to-human communication such as eye contact and hand gestures may be missing. The experts agreed on the necessity for eHMIs to support AV-to-VRU communication for safe interaction. Such devices are mounted on the exteriors of AVs or projected from them (detached HMIs) to communicate with VRUs.
Complete AV penetration will not be immediate may never reach 100%, so how will that affect the task of hazard perception and traffic interaction for VRUs? How will VRUs be able to tell the difference between AVs and manually driven vehicles? Research shows that pedestrians and cyclists want AVs to communicate with them to help them decide how to act in mixed traffic. Will it be sufficient to model communication on traditional signals, like brakes or turn indicators or will new types of signals be required?
Authors agreed that AVs should not give advice or instructions since the VRU still needs to assess the overall traffic situation to make appropriate and safe decisions. Overconfidence in the AVs message while missing other important traffic cues could result in the wrong decision. For example, if an AV signals it will stop at a crosswalk while other drivers proceed through without noticing the pedestrian. Similarly, a personalized message sent by an AV could be received by the wrong individual, causing them to take the wrong action. AV behaviour and communication protocols may also need to take into account cultural differences and regional norms for traffic participation. The information from an AVs should not be directed to just one traffic participant, and ideally would be perceivable and interpreted by other traffic participants. Current forms of communication between drivers and VRUs should somehow be reproduced, with the strongest and most intuitive cues being implicit vehicle motion - speed and direction changes and distance.
Whatever modalities are used, message content must be perceived and understood quickly and easily - as well as cutting across diversity in culture, language, age, cognitive and sensory-motor capacity. The following criteria for eHMI communication were highlighted: Cues should be simple - easy enough for children to understand, and thus avoiding the need for multiple message types to different demographics. eHMI communication must not create confusion or increase cognitive and attentional load for VRUs. However, regarding this last point, one expert emphasized that after some exposure, people will quickly learn the new AV behaviours and messages. The necessity for the universality of eHMI communications and signals requires that they not be language-dependent, but subscribe to universality across cultures, ages, literacy levels, and should be standardized. Messages and warnings could be conveyed through auditory, visual, even tactile means. eHMIs may need to be multi-modal to address individual sensory limitations as in the elderly and visual or hearing impaired. Text should be avoided, not only for the necessity for language translation, but also because interpretation is less immediate and legibility may be compromised by distance. Messages could be combined with wearable systems (e.g., smart devices carried by VRUs). Detached eHMIs, as in road surface projections, were not favoured since the view could be easily obstructed and would not likely reach all relevant traffic participants.
Will infrastructure be used to facilitate communication between AVs and VRUs?
Experts agreed that smart infrastructure will be part of the future mixed traffic solution, enhancing decision making by providing missing or necessary information. Intelligent infrastructure, as part of a C-ITS network, would collect and share information from vehicles and VRUs’ devices and facilitated through the new 5G network for real-time accuracy. Such shared information would increase the perceptual capabilities of AVs on VRU positions and movements, and provide warnings and information to VRUs. Considering the challenges to implementing this technology - cost (especially low-income countries), reliability, maintenance - it may be most applicable in areas of high pedestrian density. Some argued for AVs not being dependent on smart infrastructure. Further, such systems should not disadvantage VRUs who are not connected through wearable devices. Since mixed traffic modes are fundamentally more dangerous, where possible, VRUs and AVs should be segregated. Finally, there will continue to be a need for current infrastructure elements for individuals who wish to continue driving manual vehicles such as classic cars and motorcycles.
SAFE-UP’s contribution
SAFE-UP will be addressing AV-VRU interaction and communication in the development of new prototype technology to enhance detection of VRUs in bad weather, novel active safety functions for collision avoidance and an app through which warnings and information can be shared from connected AVs (CAVs) and connected intelligent transport systems (C-ITS) with VRUs on their smart devices.
Some responsibility must remain with road users when walking or riding bicycles and powered two-wheelers. Advanced sensors to detect VRUs are available, but prediction of pedestrian behaviour is very difficult and still under development. Until the automation and communication technologies fully mature, important intermediate solutions involve training and awareness-raising on how to discriminate between AVs & conventional vehicles, understanding AV movements and their signals, and appreciating the technological limitations and the risks of not taking these into consideration when interacting in mixed automated traffic. To meet these needs, we must be creative and establish partnerships for ongoing knowledge exchange and co-creation of effective strategies that keep pace along with the evolving traffic landscape.
