What is new in 3D and spatial computing: a field guide to the recent changes

Few technical fields change as fast as 3D and immersive technology. A method that was a research curiosity three years ago may be a standard feature of mainstream tools today, and a product that was announced with great fanfare may have faded. If you work with real places, whether in tourism, heritage, planning or training, it helps to have a clear map of what is new, what is solid and what is still unsettled.

This article surveys the main developments, explains why each matters and offers a way to decide what to adopt. It reflects the situation at the time of writing, and details such as browser support and standards move quickly, so check current status before committing.

1. Gaussian splatting and neural scene capture

The biggest change in capture is the arrival of neural and splat-based methods. Neural radiance fields, from 2020, showed that a model could learn a scene from photographs. Gaussian splatting, from 2023, made the result real-time, and tools for capturing, editing and viewing splats now exist for phones, browsers and engines.

• Why it matters: highly realistic captures of real places from ordinary video, in minutes.
• Still developing: editing, relighting, compression, surfaces for physics and a common file standard.
• Practical advice: use it for visual showcases, and keep meshes and point clouds where you need measurement or interaction.

2. Meshes keep getting better

It is easy to think that meshes are yesterday’s technology, but they have improved in important ways.

• Virtualised geometry. Modern game engines can draw scenes of astonishing triangle counts by loading only the detail the viewer can see, which reduces the need for manual simplification.
• Mesh shaders and GPU-driven pipelines make better use of graphics hardware for huge models.
• AI-assisted clean-up. Tools can now help fill holes, simplify, retopologise and generate textures, and cut hours from the work of preparing a scan.
• Methods that extract clean surfaces from neural and splat captures are an active research area, linking the two worlds.

3. AI-generated 3D

Systems that turn a text description or a single image into a 3D model have improved rapidly. They are useful for concept work, background objects, quick props and early layouts.

• Strengths: speed, low cost and creative range.
• Weaknesses: inconsistent quality, odd geometry, weak topology for animation and unclear licensing or data sources in some tools.
• Practical advice: use them for drafts and filler, check the licence of any tool and treat the output like a rough sketch that a person must refine. They do not replace capturing a real, specific place accurately.

4. WebGPU and better 3D in the browser

WebGPU is a modern way for web pages to use the graphics chip, replacing an older interface that dated from the early 2010s. It brings better performance, support for compute work and a design closer to current graphics hardware. Support has expanded across the main browsers in recent years, but not everywhere on every device, so provide a fallback.

• Why it matters: richer, faster 3D on the web, without an app, including splat viewers and large scenes.
• Practical advice: use libraries that support both the old and new interface, and test on the devices your audience actually owns.

5. Standards for sharing

• glTF is now the default way to deliver 3D on the web and in AR, with extensions for compression, materials and, increasingly, new kinds of data.
• OpenUSD, backed by an industry alliance formed in 2023, is gaining ground as a way to assemble complex scenes from many tools, particularly in film, design and industrial digital twins.
• 3D Tiles streams huge geospatial datasets, from city-scale photogrammetry to terrain, and has been extended with richer metadata.
• Work to carry Gaussian splats inside glTF aims to let splats travel like any other asset.

Standards matter because they protect you from lock-in. Prefer open, widely supported formats for anything you want to keep.

6. Mixed reality and spatial computing hardware

Headsets have moved from pure VR to devices with high-quality colour passthrough, which show the real world through cameras and let digital objects sit in your room. Hand and eye tracking are becoming standard, so you can interact without controllers. Some devices are positioned as spatial computers, aimed at work, media and design, not only games.

• Why it matters: mixed reality becomes practical for training, design review and guided work in real places.
• Still developing: price, weight, battery life, comfort over long sessions and the range of useful applications.
• Practical advice: design for several devices, and favour web-based and standards-based content, so that you are not tied to one headset.

7. Smarter capture hardware

• Phones and tablets with LiDAR sensors make quick room scans possible.
• Handheld and backpack scanners map interiors and large sites quickly.
• Cheaper drones carry sensors that once cost a fortune.
• Cameras with 360 degree coverage and better stitching speed up virtual tours.

8. Digital twins grow up

Digital twins are moving from showpieces to working tools, with better links between 3D models, sensors, GIS and analytics. The challenge has shifted from building the model to keeping it current, governed and useful, which is a people and process problem more than a graphics one.

How to judge a new technology

1. 01What real problem does it solve for us, and who would use it?
2. 02What does it cost to capture, produce and maintain, over two years?
3. 03How mature are the tools and standards, and who supports them?
4. 04What happens if it fades? Can we keep the source data and move to something else?
5. 05Can we test it cheaply first, on one place and one audience?

A simple plan to keep up

• Follow two or three reliable sources, such as standards bodies, tool release notes and respected researchers.
• Set aside a few hours a quarter for a small experiment with one new method.
• Keep your raw data, so that you can re-process it when better methods arrive.
• Share what you learn with your team in a short note.

Frequently asked questions

Should we switch everything to Gaussian splats?

No. Use them where realism matters most, and keep meshes and point clouds where you need to measure, edit or interact.

Will AI replace 3D artists and surveyors?

It is changing their tools and speeding up parts of the work, but accuracy, judgement and knowledge of the place remain human responsibilities.

Is it worth buying a mixed reality headset now?

For testing and for specific training or design uses, yes. For broad public reach, web-based 3D and phone AR are still the safer choice.

How do I protect my project from changing technology?

Keep original captures, use open formats and separate your content from any one platform.

The pace of change is a reason for curiosity, not for anxiety. Understand what each development is for, test it on a real problem and keep your source material safe, and you can take advantage of what is new without being swept along by it.