CityEngine: Build a Digital Twin of Your City | Hacker News Reader

CityEngine: Build a Digital Twin of Your City

gisgeography.com

119 pointsrntn57 comments

Has anyone here played the new Microsoft Flight Simulator Game they released in 2020? I was blown away by how accurate the environment and maps were (at least in Chicago and Bloomington, IL). By far the most realistic representation I've seen in a game before. Even outside of the iconic buildings of downtown, a lot of the smaller buildings in far out neighborhoods were more realistic than I was expecting.

I know the game has gotten a lot of praise on the graphics. I guess they leaned heavily on bing maps to build out the game's environment. Would be crazy to see what could be created with something like google streetview data.

Bing gets the credit, but much of the data is from OSM.

Indeed, the Melbourne Monolith for example, for which I love the idea.

https://australianaviation.com.au/2020/08/typo-caused-melbou...

The thing I find strange about MSFS is that the 3D visualisation is in many cases much worse than the Google/Apple maps. It seems to replace many buildings with generic models - they have the same shape, but the textures and style is obviously just a generic sculpted to fit. Apple/Google maps use the actual textures of the buildings themselves.

For my home city for example Apple maps has full 3d for every single building and it looks amazaing, but in MSFS most of the buildings are replaced with generic copy paste American suburbia style structures.

It's likely to help keep the size of the game data "under control" vs the map data streamed at you by Apple Maps or Google Earth / Street View bein' streamed in small(ish) chunks from a massive database of texture data stored on corporate server disk pools.

The game actually streams in much of its contents also. They still stream in the satellite ground textures, but they don't use these to texture models in the same way as Google/Apple maps.

You get much a much higher LOD in the simulator than on Google; gaming GPUs have a lot more to work with than integrated/consumer GPUs. Google Earth just looks better sometimes because the source data and reconstruction is better. They are really dedicated to in-house photogrammetry for many years, when Microsoft/Bing have just been buying data from 3rd parties.

I'm over 150hrs into it and an obsessed. I mostly fly in the Bay Area. The data is old but is good enough where I could not only find my house, but saw my car in the driveway.

There's a free plugin for swapping in Google maps data to but it's really best for areas without photogrammetry.

But the sim is so detailed that when I went on a discovery flight IRL I had zero issues finding my home from the air because I already was familiar with the view from up there.

I had the opposite experience, I was disappointed to see how some famous cities (Copenhagen and Rio de Janeiro) look really bad with generic buildings. Maybe it's a US vs rest of the world thing, or am I missing any setting here?

Could be your bandwidth or not having the version up to date. Living in Copenhagen, it is remarkably accurate. I literally crashed into my apartment.

Only very vaguely related pet peeve: the term "Digital Twin" implies that a simulation perfectly matches what is being simulated. This is really almost never the case.

Let's just call them what they are: simulations.

Digital twins aren't simulations, they're just data. The idea is that it's openly available and easily queryable so that people have good data to run their models and simulations on.

Even real life twins don’t exactly match. No digitization process can be error free, the “digital twin” language is to indicate that it’s supposed to be a realistic model rather than an editorialized/interpreted environment.

But a realistic model for what? The question of what data is important to capture, and how, depends entirely on what you're trying to do with it. The danger of the digital twin narrative is the idea that one model can serve all purposes, which I think will ultimately lead to broken models.

...addendum: ALL models are editorialized/interpreted. Forgetting this often leads to errors.

It's a shame that it requires a business email just to try it out. My work doesn't have anything to do with city building but I'd love to use this to create a 3D model of my home town for my own fun.

There's a blender plugin that will let you import 3d map data including buildings and trees from google into blender and then do whatever you want with.

https://github.com/domlysz/BlenderGIS https://github.com/eliemichel/MapsModelsImporter

Hey, hey, this looks great! Thanks!

You can use any email address that's not gmail, icloud, yahoo, et al.

Just register a cheap domain name and point its DNS records at a free email provider (or setup an incoming mail server... outgoing with getting labeled spam is the hard part anyways)

Good to know. Unfortunately, my business email IS a gmail address.

If you'll allow a shameless plug: I'm about to publish a python package to help doing just that, by combining photogrammetric projects.

The workflow is take a couple hundred overlapping pictures, compute a model using an existing backend and store the image/3d data. Rinse and repeat. When you have a few neighbouring projects, we recompute submodels at the boundary and merge the involved 3d models by reprojection.

Currently wrangling some issues around packaging, a BSD licensed functional prototype should be out sometime next week. I'd love to get some users or eyes on it, do get in touch!

Stay away from ESRI (ess-ray) - I should get that printed on a t-shirt.

ESRI centralise power, lock you into their technology, and IMHO a digital twin should be doing the opposite. Don't believe the sales BS about ESRI having "free" software, read the fine print, ArcGIS AINT FREE.

IMHO a Digital Twin should be something for the people, by the people! So please stop paying proprietary systems like ESRI and incentivising them to stifle innovation and increases barriers to entry.

If you want to see local jobs, growth and innovation then invest in OpenSource software like OSM (Open Street Map) and QGIS and other Open GIS products.

I have worked with a few people with GIS degrees and 99% of them know ArcGIS and ArcGIS only. This situation reminds of how dominant SAS and Stata used to be before R and Python exploded.

Is there a FLOSS GIS solution that could rival ArcGIS?

Yes and no.

The problem is that ArcGIS isn't just software; it's also data, data management/storage/interop, process automation, and so on.

There are other sources for the data, and a lot of the data is open, but the Python/R model doesn't translate directly because those sources have to be individually licensed and integrated. And once in a blue moon you'll need ArcGIS for the data anyways and now your whole FOSS investment is up in flames because you fork over the license to get the data anyways...

You can do a lot of what ArcGIS can do if you know the entire ecosystem and all the data sources and how to glue them all together. But that skillset approaches "Senior SWE", at which point the generic GIS major + an ArcGIS license is waaaaay cheaper.

In fact, ESRI could probably get away with drawing even more blood.

(don't get me wrong; I think OP is correct that esp gov't agencies should invest in FOSS solutions, and the amount of effort required for a full digital twin is definitely worth NOT doing on top of ESRI tech... but the value add of ArcGIS for most GIS applications is difficult to deny given the differential in labor costs you get by staying in the ESRI sandbox... there are literally hundreds of undergraduate degree programs that basically exist to churn out affordable ArcGIS techs... the same is not true for FOSS alternatives, where to get comparable expertise you literally need to hire something comparable to a Senior SWE; it's going to cost you waaay more than "fresh GIS major + $4K".)

I’m far from a GIS expert, but there’s QGIS: https://www.qgis.org/

Esri is becoming like the Microsoft Office/SharePoint/Teams/Power ecosystem. Many of the individual products are of limited quality and better alternatives exist. But taken together they have network effects and are easier for corporations to manage and justify.

The open solution is QGIS and PostGIS. They are ridiculously capable. But I think the next battleground is actually access to data rather than the software. Esri have been building an open data repository that is becoming favoured by a lot of government organisations. They stream data to esri users and make people reliant on the cloud service. And add annoying intermediate steps for people wanting raw data suitable for loading into a real database.

We should ask the long term question, what if QGIS and other FLOSS products had broad support and significant resources flowing into them that ESRI has?

The way I think about it, often the same people, certainly people with the similar skillsets are working on all these products, most of them just follow the money.

I think one big missing piece of the puzzle for governments to assess these products is a governance and risk model associated with these products. But software engineers generally don't give a crap about governance, so we need more people who do care about that to invest in OSS. I'm seeing some movement on this like the Github Sponsors initiative.

+1 for QGIS and PostGIS. Without being a GIS expert I was able to develop a solution for storing and querying invasive species abundance (and other ecological) data using these tools, and they worked _really_ well, and the researchers who were using that data were able to access it from their R environments without issues.

I know an MS GIS who uses Arc about once a year now, mostly Q, R, and Python.

I am not a GIS expert, but I work with spatial data all the time and have never touched an ESRI product (besides pulling data from servers running their crap). I do nearly everything in QGIS and publish using GeoServer or standalone PostGIS. I have never encountered a situation that would be only possible or way easier to use ArcGIS.

I feel like (and I'm welcome to being corrected by any experts here) ESRI stuff is kind of like Adobe CC - the vast majority of people have absolutely no need for Photoshop/Premiere/... and could do everything they need just as easily with any one of the many open source or cheaper alternatives. Most people use it because that's what they learned on, which is because that's what most people used at the time (and also because of very "generous" education licenses, teacher training...)

100%. Their stronghold on governments (at all levels -- municipal, state, public safety/emergency response) is almost total and quite unfortunate. Very reminiscent of Microsoft/Windows/Office (which have strongholds in the exact same places) or Oracle.

As one example, ESRI's web mapping offerings are very behind the times (in terms of usability, api's, etc.) Mobile support has been atrocious. Outdated software practices + government lock in = terrible public maps and little innovation.

It's not a coincidence that almost nobody chooses them in the private sector for their maps. Just as you'd be surprised to learn that a startup picked Oracle as its database vendor...

QGIS for life

Here's they have an example from Portland, OR. Seems.... pretty basic I guess? https://www.arcgis.com/apps/CEWebViewer/viewer.html?3dWebSce...

Not sure the pricing.

yeah I'm not getting the product other than it looks it ties into arc gis

For a comparison, here is google earths version of Portland (this isn't and apples to apples comparison, but I still think valid)

https://earth.google.com/web/search/port+land/@45.514524,-12...

It's not that the esri one is bad. It's that the GE one is amazing.

I am constantly having my mind blown by Google Earth. How can this be free?

It's possible to download Google Earth's 3D model, at the resolution in your link, for any area where they show 3D buildings/landscape. It's not something they facilitate and definitely requires something of a hack, but it does work. Then you can get the model into Blender, SketchUp, etc. and start editing it to add or remove buildings and so on.

Would be interested to know whether that's technically legal. I mean, they put it out there on their website, right?

NAL, but surely they put something in the ToS that says you aren't allowed to do that. And even if you're 100% sure it's unenforceable, are you willing to defend your company against Google's army of lawyers? Scraping data from the web may be legal in many places, but that says nothing about what you're allowed to do with that data (hint: usually not much unless it's your own user data).

Having said all that, I've done this in the past for a few side projects where I wasn't selling anything and knew they would never in a million years get big enough to attract Google's attention. For those, this is definitely a good way to get 3D data (along with OSM Buldings, if you don't care about realism and just want a good approximation)

Shame there’s no free trial. I guess I’ll stick to Cities: Skylines for now.

A quick Googling turned up a Cities:Skylines map importer for real world terrain:

https://www.pcinvasion.com/how-to-import-maps-cities-skyline...

One idea I had was to take my home city and improve it with more pedestrian streets, greenery and light rail and to see the effect on traffic

Sounds fun! Feel free to share the results :)

Does anyone know what the limitations are for the free trial? I'd love to try this with where I live and experiment with it in Unreal, and pay for a license if I do anything commercial with it.

Not sure about the trial.

It’s included with ArcGIS for personal use now, which is $99/year and includes the rest of the ArcGIS system (except enterprise server and the like). Obviously for personal use only.

We also released our maps SDK for unity and unreal: https://developers.arcgis.com/unity/

Disclosure: I work at Esri but not in a sales role. I am also a customer.

Shameless plug: I'm working on a project in this space (see my other comment on this post), I'd love to hear about your prefered workflow for ingesting 3d data into your editor.

Put another way, what output formats are most convenient for your use case? Are raw point clouds (XYZRGB) useful to you for eg. "prototyping volumes" or do you require meshes as a starting point? If the latter, how would you quantify how much retouching is acceptable (holes, suboptimal meshing, irregularities...) for an urban model?

Anyone know of any good use cases for a 3D models or simulations at the city scale apart from eye candy, e.g. public presentations of future urban development?

I worked in this space for a while and I never found much of a use case for serious applications of 3D over say an ordinary 2D GIS. Ended up building applications using PostGIS openstreetmap etc that were a lot more effective.

Knowing building volumes and surface areas can be really helpful for studying Urban Heat Islands since they give you a more accurate idea of the amount of impervious material/surfaces absorbing and emitting heat.

It's very useful in film CGI, for example if you need a background for a flying helicopter/plane/superhero or, if it's detailed enough, need to simulate a building collapsing or even just the shadows of a flying object to then overlay on your real footage. Even if you need to manually model the block you're filming on, it's still a great starting point and can be used for anything far enough or not in focus.

Outside of that, I know people also use it to plan solar panel installations (simulating shadows to optimize roof solar panel placement) and I'm planning to use it for determining how much sunlight and subsequently heat different streets get with the eventual goal of being able to route around the hottest streets when walking in the summer.

I like the idea but wonder why they have chose such a blurry basemap, it makes it look very very awful, like an early 2000s product.

Wouldn't it be a more compelling product if you sold the model, not the model-making tool? If I was the city of Portland, I would rather just buy the pre-made Portland model than have an intern build it with CityEngine.

CityEngine is a procedural generation tool, it isn’t really meant to compete with real world data providers. For something like Google Earth but available as GIS data, you can use a provider like NearMap.

CityEngine is better for planning out and comparing future scenarios, and other specific use cases. E.g. https://m.youtube.com/watch?v=sOY3LY688QI

Disclosure: I work at Esri but not in a sales role. I am also a customer.

My SimCity nostalgia wishes someone would use this to make agent based models of real cities with every citizen mapped out.

Anything like that ever been done?

Have you played Cities: Skylines? There are definitely limitations to its fidelity, but every person and car is something you can click on and see their home, job, route. This gives much more realistic traffic for one thing (though there are mods that do an even better job).

Ultimately though, I doubt that high fidelity simulations are in the cards for consumer devices any time soon. Cities Skylines already bogs down on large cities.

MATSim or ActivitySim are used for that sort of thing. Most modellers don't bother to simulate every individual person though. Instead they use a sample which they scale up because it isn't worth the order of magnitude more computing time when just simulating a representative 10% does fine.

https://en.wikipedia.org/wiki/Stasi

Wow, this is the utterly most uninteresting aspect of city planning. But good that that problem is solved.

Looking forward to seeing this implemented in 5 or 10 years at an exhibit at a regional small-to-medium size science museum.

Your move, SF YIMBY.

Reply on news.ycombinator.com