• Make ADS vehicle movements more “readable” and consistent with existing road user expectations. Variation in vehicle driving styles may lead to difficulties in interpreting the intention of the driver behind the vehicle. ADS vehicles have the potential to create more standardized vehicle-movement cues that are easily interpretable. 
• Replace traditional driver-related cues with vehicle-signalizing systems such as eHMI to better communicate intent with vulnerable road users. As pedestrians and cyclists currently rely more on driver-related cues over vehicle-signaling systems, it is important for ADS vehicles to assist them in identifying the intent without verbal communication or hand signals, since future ADS vehicles will not have a driver to provide such cues. 
• Design eHMI that clearly and intuitively indicate intended vehicle action or suggested pedestrian action for highest comprehension. Training on the meaning of interfaces helped further enhance comprehension of all eHMIs. However, interfaces that were unfamiliar and unintuitive were unpopular, even after the meanings of the interfaces were explained. For example, the "don’t-walk" and "walk" icons were the most preferred interface for participants as these icons are universally recognizable.
• Design ADS vehicles with eHMI to interact with shared road users such as bicyclists that may be in close proximity. Driver related cues are used mostly when vehicles are in close proximity to shared road users, for example bicyclists who are concerned about being seen by nearby drivers often rely on eye contact and hand gestures. eHMI should be designed for such situations.
• Notify ‘yielding’ information as early as possible. It was found that participants could better comprehend the intentions of a yielding vehicle if the vehicle’s eHMI changed to its final state when it began deaccelerating rather than when it reaches a full stop. Participant feedback also suggested that earlier notification was preferred.
• Avoid binary options with confidence scales if using a survey approach to assess confidence perception of eHMI. Overconfidence ratings were found in this study as only bimodal options were provided. One potential way to address this overconfidence in future research is to provide participants with three distinct response options (i.e., yes, no, don’t know). Moreover, it is recommended to provide instructions that clearly and objectively delineate the different response options.

It is worth noting that the study results showed promise in potentially differentiating between different eHMIs for the single specific scenario involving a relatively simple crosswalk case; however, the generalizability of findings was not claimed.