Objective Verification Method for “Unsafe Pedestrian Access” Assessments Using Gemini AI

General Discussion
1

This time, I conducted an objective and reproducible verification regarding Wayfarer’s “unsafe pedestrian access” assessments using Gemini AI.

Rather than relying solely on subjective explanations, I designed the process so that third parties could independently verify pedestrian accessibility conditions using only the provided videos.

The following is the actual on-site verification process I performed.

【On-Site Verification Process】

・Selected a Google Street View-supported location as the starting point
・Prepared two smartphones
・Used one device for on-site video recording
・Used the other device for GPS observation with Google Maps running
・Started video recording simultaneously on both devices
・Maintained real-time GPS position updates throughout the process
・Actually walked to the destination on foot
・Paid attention to safety and privacy during recording
・Naturally recorded road conditions, terrain, fences, and surrounding environments
・Slowly rotated the camera at notable locations
・Ensured that the GPS-recording video simultaneously captured the same rotational movement in sync
・Avoided editing or unnatural interruptions as much as possible and maintained continuous recording
・After arriving at the destination, recorded the Wayspot candidate and surrounding conditions in detail
・Saved both videos
・Uploaded both the walking video and GPS video to YouTube or similar platforms

This verification method focuses on the following points:

“Demonstrating that the movement was actually performed on foot”
“Showing synchronization between GPS position data and on-site video footage”
“Allowing third parties to independently evaluate the surrounding environment”

After that, I entered the following verification prompt into Gemini AI.

━━━━━━━━━━━━━━━━━━
【Prompt Entered into Gemini – Start】
━━━━━━━━━━━━━━━━━━

Please analyze the following two videos for route continuity, terrain consistency, GPS synchronization, and pedestrian accessibility.

【Video URLs】

Walking route video:

https://www.youtube.com/watch?v=_35mVnfxW8U

GPS-synced route video:

https://www.youtube.com/watch?v=ODdn8yNWy9E

【Items to Verify】

■ Route Continuity

  • Check whether the walking route appears continuous.
  • Check for unnatural cuts, jumps, warping, or editing artifacts.
  • Determine whether the movement speed and camera behavior are consistent with normal walking.

■ GPS and Terrain Consistency

Please verify whether the following elements are consistent with the video footage.

  • GPS movement
  • terrain and elevation
  • road shape
  • embankment structure
  • pond/reservoir layout
  • surrounding structures and fences

■ Timeline Analysis

Please summarize the environmental changes approximately every 1 minute.

Example:

0:00–1:00
1:00–2:00
2:00–3:00

For each section, briefly describe:

  • current location status
  • surrounding environment
  • approach progress
  • pedestrian accessibility

■ Pedestrian Accessibility and Safety Evaluation

Please objectively analyze whether the route appears walkable and reasonably safe under normal daytime conditions.

Please evaluate:

  • whether normal pedestrian access appears possible
  • whether the route width is sufficient for walking
  • whether the terrain appears dangerous
  • whether there are obvious hazards, collapses, or restricted areas

Also include a final “Safety Level” evaluation using one of the following categories:

  • Extremely Safe
  • Generally Safe
  • Neutral / Ordinary
  • Somewhat Dangerous
  • Dangerous
  • Extremely Dangerous

Please explain:

  • why that safety level was selected
  • which visible elements in the footage support that conclusion

■ Cross-video Consistency

Please determine whether the two videos appear to depict:

  • the same location
  • the same route
  • the same target area

【Important】

Please prioritize objective analysis based on:

  • visible evidence
  • GPS behavior
  • terrain
  • structures
  • road conditions
  • walking continuity

Please clearly separate:

  • objectively verifiable facts

from

  • assumptions or speculation

Please respond entirely in English.

━━━━━━━━━━━━━━━━━━
【Prompt Entered into Gemini – End】
━━━━━━━━━━━━━━━━━━

※ Please replace the uploaded video URLs with your own before using this template.

Gemini’s analysis produced the following conclusions:

  • Continuous pedestrian movement on a paved route was confirmed
  • High consistency between GPS movement and terrain
  • No visible “No Trespassing” signs
  • No major collapses or dangerous structural hazards
  • General pedestrian access appeared possible
  • Final safety level: “Generally Safe”

Of course, I do not consider AI analysis to be absolute.

However, by combining:

  • actual on-foot recordings
  • GPS synchronization
  • publicly available videos
  • the published prompt
  • the ability for third parties to independently reproduce the verification

I believe this may serve as one example of a more objective method for evaluating pedestrian accessibility, rather than relying primarily on subjective impressions alone.

I hope this can serve as a useful reference for future rural Wayspot nominations and areas where Street View coverage is limited.

This verification method was developed during discussions related to a rural nomination that received an “unsafe pedestrian access” assessment:

2

Hello @Nena
That is for putting this all together.
You have clearly worked hard to find this methodology.
I appreciate that it has been inspired by the issues you have encountered with a specific nomination, and the rejection reason around safe pedestrian access.
Wayfarer is all about find specific points of interest so each actual location/POI is unique. It is then about making decisions based on guidelines and points of principle.
There will always therefore be a degree of flexibility depending on that particular POI.
I am very happy that the decision comes down to human interpretation. I might not always agree with it and others may not always agree with my decisions about a POI but that is the nature of wayfarer. And I have to accept that.
You have set out your case for this particular POI as best you can. Well done on the effort. But there comes a point where you have accept that views are going to differ.

3

In particular, you cited “No Trespassing” signs. However, wayspots can be legitimately placed beyond such, because the standard is safe pedestrian access for those authorized to access the property; public access is not required.

For example, in my community, there is a wayspot located in an employees only hallway inside a courthouse, and two located inside buildings at a transit district vehicle maintenance facility.

I’ve also been to two zoos and two amusement parks and all four are loaded with wayspots despite requiring paid attendance.

4

Hello @elijustrying

That place is a quiet reservoir hidden deep in the mountains, and it is truly a wonderful place.
At this time of year, I imagine fireflies must be dancing through the air from dusk onward.
It reminds me fondly of long ago, when I used to go bass fishing there in the evenings.
(Fishing is now prohibited there.)

Like you, I also believe that human judgment is extremely important.

I am not in a position like yours, and I am still quite inexperienced in many respects, including Wayfarer knowledge, experience, and communication skills.
And I assume that you yourself have already reviewed the links and details regarding my nomination.

You mentioned that:

“I am very happy that the decision comes down to human interpretation.”
and
“There comes a point where you have to accept that views are going to differ.”

So I would genuinely like to ask:
if you were to step away from your position as an Ambassador and review my nomination simply as a person, would you personally lean more toward “accept” or “reject”?

I would truly appreciate hearing your thoughts and impressions.

5

Hello @CostcoShopper13
Are you perhaps a player from an urban area?
I am a rural player.

I am genuinely happy that discussions like this allow us to learn about many different types of Wayspots around the world.

Here in rural areas, there are many locations where access can become temporarily restricted depending on conditions. For example, shrines where roads freeze in winter, seaside shrines that become inaccessible during high tide or rough waves, and even shrines on small uninhabited islands located about 500 meters offshore, reachable only by swimming or by boat. I have personally visited such an island twice by local transport boat.

In my case, I have repeatedly experienced nomination rejections and appeal rejections. What has been especially difficult is trying to prove that these places are not actually dangerous.

My goal is not simply to get approvals.
What I truly want to understand is why they are being rejected under the reason:
“Unsafe pedestrian access or possibly not publicly accessible.”

I believe reviewers should also be careful to remain fair toward players from many different environments and regions around the world.

6

If any Wayfinders are interested, I would genuinely appreciate it if people could try experimenting with the prompts themselves by slightly changing the wording, using different locations, or testing different descriptions to observe how AI responses and interpretations may vary.

For example:

・Changing the phrasing slightly
・Replacing videos or images
・Changing URLs
・Actually visiting and verifying locations in person

By changing different conditions, it may become possible to better understand how AI interpretation and responses can change.

I do not believe that AI is always correct.
Rather, through this case, I hope the community can objectively explore and discuss questions such as:

“What is AI using as the basis for its judgments?”
“How does AI interpretation differ from human interpretation?”

I believe that reproducible experiments and open discussion may help us better understand where misunderstandings or inconsistencies can occur.

7

If you are talking about the ML (machine learning) system that Niantic use to filter submissions, any attempt in this forum to collaboratively dissect how it decides, in order to work around it, would be inadvisable.

8

Hello @salixsorbus
I had not been actively involved with Wayfarer for quite a long time.
Recently, I casually started submitting nominations again.

I honestly do not know much about the existence of ML systems.
My reaction was simply,
“Oh, systems like that have been developed? That is quite impressive.”

The Ambassador stated:

“I am very happy that the decision comes down to human interpretation.”

Because of that statement, my understanding has been that the final decision ultimately belongs to human reviewers.

I am very grateful to the people who gave me various pieces of advice regarding the rejection reasons for my nomination.

After receiving rejections and continuing to improve my submissions based on the advice I received, I gradually began wondering:

“Could AI perhaps be used as an objective aid for verifying pedestrian safety?”

As far as I understand, this prompt is intended only to help objectively document and verify real-world on-site safety conditions.

If someone were attempting to use such methods for the purpose of bypassing an ML system, I believe that would constitute abuse of the system.
Such behavior should not be acceptable.

I would actually encourage people to try simple personal experiments themselves using two smartphones for safety verification purposes.

I do not believe there is anything wrong with individuals personally testing, observing, and enjoying the results for themselves.

Thank you for taking the time to express your concern.

9

The ML system is currently used to remove garbage submissions from the wayspot review queue. Without it, the review process would be unpleasant, as the garbage would be overwhelming.

It cannot review and determine every aspect of a submission, which is why it does not accept wayspots, but only rejects them.

10

This statement troubles me. Testing what you can get away with is not something we should do as wayfarers. We should be submitting wayspots that we honestly believe meet criteria as stated here:
https://niantic.helpshift.com/hc/en/21-wayfarer/section/166-wayspot-criteria/

11

I do not understand why we need to use AI to state whether a location is safe.

  • Most locations are obviously safe, they are not going to put a play park in a dangerous location etc.

  • If the location looks like it could be unsafe then it is down to the submitter to realise this and show evidence to the fact. If this is not supplied then the reviewer should Reject.

It’s quite simple really and Rural / Urban does not change this.

12

@cyndiepooh
@SlimboyFat71

Please read the original post and linked material first.

If, after reviewing the videos, GPS verification, 360 images, and the actual evidence provided, you still do not understand the point being made, then we are probably discussing two completely different problems.

13

I have attempted to read both of these posts, and I admit that I have no idea what you are trying to do here. You want to add an extra ai step to determining safe pedestrian access? You want to have the automated process automatically overturn safe pedestrian access issues? I can’t follow your walls of text and see no need to complicate this with ai.

14

All I can see from your OP is that you want to make the reviewing for “Safe Access” much harder than it needs to be.

If you are getting “Repeated Rejections” when many others are not then you need to be looking at why that is, what is missing from your nominations that mean that reviewers can not confirm “Safe Access”.

Remember that the onus is on the submitter to make sure the reviewers has everything needed to be able to review the nomination.

15

I am, yes. I am in the San Diego, California metro area. However I formerly lived in a nearby mountainous rural area, so I do not see things purely from an urban perspective.

But the point I was making is generally applicable everywhere as it is based upon the Wayfarer standards.

Simply put, a “No Trespassing” sign or any other kind of restricted access, does NOT disqualify a location. Wayspots are permitted as long as there is physically safe pedestrian access, even when it is not available to the general public (my examples being government agency buildings that are restricted to employees only) or requires a membership or an admission fee. With the categorical exclusion of schools and single family residences, but that is based on their identity, not accessibility.

I do understand what you are getting at, I am just pointing out that your suggestion regarding access signage is based upon a misunderstanding of the applicable Wayfarer standards.

I have had quite a few rejections myself, but they are usually based either on the content of the nominations, or reviewers misunderstanding and misapplying rejection reasons (such as rejecting apartment community swimming pools as being on single family residential property).

16

@CostcoShopper13
Thank you very much for your detailed explanation.

My explanation regarding access may have been unclear, and I apologize for that.

I understand that access restrictions themselves do not automatically disqualify a Wayspot nomination, as long as there is physically safe pedestrian access.

What I was trying to explain is that some reviewers may misunderstand rural reservoir areas or locations that appear to have restricted access as having “unsafe pedestrian access,” even when they are actually walkable and safe under normal conditions.

There may also be differences in perception between urban and rural reviewers.

Because of this, I began documenting actual access conditions in greater detail through synchronized GPS walking videos and AI-based comparison tests.

I have also experienced several rejections in the past that I believe may have resulted from reviewer misunderstandings.
For example, I feel that my explanation was insufficient in a nomination for a specialty shaved ice shop.