Designing human-like deceleration to improve yielding communication for automated vehicles.

For Autonomous Vehicles (AVs) to operate safely in mixed traffic situations, they must be designed to be understood by surrounding road users. To replace the communication with onboard drivers', external Human-Machine Interfaces (eHMIs) were designed to communicate yielding intention. Indeed, many recent studies have focused on designing these eHMIs and investigating how these eHMIs affect pedestrians' crossing behaviour and experience.

Numerous benefits were shown from these eHMIs; such that pedestrians were more likely to cross in front of the AVs with eHMIs and higher acceptance and trust towards the AVs. However, using the University of Leeds CAVE-based pedestrian simulators (HIKER), we have demonstrated that pedestrians could over-rely on the eHMIs, which leads to unsafe crossing, as demonstrated by 30% of the pedestrians crossing in front of a non-yielding car, leading to crashes (Kaleefathullah et al., 2020).

This finding expresses a pressing need for a more effective resolution for AVs to communicate intention.

Research gap and objective: One could argue that these eHMIs were potentially found beneficial, and pedestrians were overlying them because the AVs were not providing good communication via the kinematics channel because they were not presenting a human-like deceleration to communicate intention. In fact, in real-world observations, we found that conventional drivers rarely use explicit communication (i.e., hand waves, flashing headlights, or honking) to communicate with other road users (Lee et al., 2020).

Given that eHMIs could lead to potentially dangerous situations, it is time to revisit the solutions. One of these could be designing a human-like deceleration, which could improve pedestrians' understanding of the AVs' intention and thus help make safe judgments and increase acceptance and trust towards AVs. This EPSRC-DTP student project would close this research gap by designing human-like deceleration to improve yielding communication for AVs.

Method: The student will make use of the previous data set from COMMOTIONS, a previous project led by the co-supervisor Professor Markkula, and also the MIT naturalistic data set provided by our project collaborator Dr Josh Domeyer from Toyota Motor North America Research & Development. These data sets will be used to model human-like deceleration.

The collaboration with Dr Domeyer could also lead to a more long-term engagement. Different deceleration profiles (i.e., for different speeds, time gaps and scenarios) will then be evaluated from different aspects, such as pedestrians' ability and how fast it takes them to understand AVs' intention. The few most effective deceleration profiles will then be implemented in our world-largest, CAVE-based pedestrian simulator (HIKER), in which we will investigate pedestrians' crossing behaviour and experience in a safe and controlled environment.

We also develop our own simulation software for a Virtual Environment (VE), using a team of dedicated software developers, providing flexibility and uniqueness in scenario design. This project will ensure inclusive design in AVs by also evaluating and thus designing for pedestrians of all ages. Ageing causes degradation in perceived abilities and motor controls due to ageing (European Commission, 2022), leading to unsafe crossing judgments, yet a massive research gap needs closing.