FS Studio Friday DigiTalk: SlamBox, Scene Collision, GenAI and Simulation, Reality Capture, and Conferences!

Hey everyone, happy Friday!

We’ve had a packed week of exploration, innovation, and a few surprising moments of nostalgia. Let’s jump right in.

Meet SlamBox: Turning the Physical World Into a Smart, Searchable Digital Twin

Let's kick this off with something we are really excited about, a groundbreaking new tool called SlamBox, a real-time, AI-powered spatial platform that brings together 3D environments, computer vision, and large language models to transform how we interact with physical spaces.

SlamBox allows us to create richly detailed digital twins of real-world environments using nothing more than a mobile phone and a camera. Once mapped, these environments become fully interactive, we can track people, vehicles, or equipment in real-time and query the scene like a search engine. Want to know how many toilets are in the building? Curious if someone without a red hard hat entered a restricted zone after 5pm? SlamBox can answer, and act, in real-time, thanks to seamless integration with LLMs and smart sensors.

But this isn’t just about data tracking. SlamBox unlocks new ways to explore and manage environments with tools like real-time measurement, annotation, lighting edits, heat mapping, and semantic search. From facility management, retail planning, and training, to safety compliance, AR inspections, and emergency response, SlamBox brings powerful intelligence into the physical world. Think of it as the missing link between your digital strategy and the physical reality around you.

Here are some examples of how SlamBox can elevate your operations:

Retail

• Understand Customer Behavior Track foot traffic, dwell time, and movement patterns. See where customers stop, what they look at, and how they navigate your store—then turn that into actionable layout and product placement decisions.

• Real-Time Queries on Your Environment Ask things like: “How many people visited the dairy aisle after 5pm?” or “Which display gets the most attention on weekends?” SlamBox delivers fast, accurate responses using spatial data + LLM intelligence.

• Heatmapping and Zone Insights Generate heatmaps to understand high-traffic and underutilized zones. Use this data to drive promotions, reposition inventory, or streamline staffing strategies.

• Improve Customer Safety and Compliance Monitor restricted areas, trigger alerts if PPE is missing, or guide customers safely in emergencies through AI-driven real-time visualizations.

• Boost Inventory Planning and Store Design Use high-resolution 3D scans to test new displays, simulate holiday traffic flow, or refine endcap positioning—all before making physical changes.

• Scalable, Cost-Effective & Camera Agnostic Use any camera—including security cams or webcams—to bring SlamBox to life in your environment without major infrastructure changes.

Warehouse

• Interactive Digital Twin + AI: Model your facilities and track people, vehicles, and activity in real-time using any camera, paired with a mobile-scanned environment.

• Natural Language Querying: Ask questions like “How many fire extinguishers are in this building?” or “Alert me if someone enters this area after 5pm,” and get real-time answers from your environment.

• Powerful Tools for Action: Take measurements, simulate changes, annotate environments, and visualize updates – all inside a 3D digital twin.

• LLM-Powered Intelligence: Go beyond static maps. Get semantic insights and pattern analysis across space and time – from foot traffic heatmaps to object detection and alerts.

• Safety, Training & Inspection Ready: Use for AR-guided inspections, safety compliance (PPE alerts, exit guidance), and immersive training simulations.

• Flexible & Scalable: Industry-agnostic, cost-effective, and scalable across retail, construction, logistics, security, and beyond

Of course there are many other ways SlamBox can help your organization no matter what industry you are in.

SlamBox is now available for pilot engagements, offering a full 3D scan, remote annotation, and LLM-powered querying to show how your space can become smarter, safer, and more intuitive. Reach out to Bobby Carlton to learn more and get started with a pilot project.

Apple Vision Pro Scene Collision

This week, we got a look at integrated scene collision running on the Apple Vision Pro, and it’s safe to say we’re seriously impressed. When virtual objects not only appear in the real world but react to it, bouncing off walls, colliding with tables, sliding across real floors, it’s a whole new level of immersion. Beyond just being visually stunning, this has massive real-world applications. Imagine technicians learning to repair machinery with true-to-scale virtual components that behave like the real thing, but without the risk of breaking anything. Or architects and builders walking through digital models on actual construction sites, seeing exactly how a beam adjustment might cause a virtual collision before ever lifting a finger.

Even in areas like vehicle safety, this approach opens new doors, allowing for crash and pedestrian simulations layered onto real-world environments to help build smarter cities. Retail could get a boost too, think about placing a new couch in your living room and watching it react to your space in real time. And of course, gaming gets even wilder: mixed-reality escape rooms, tactical games, and educational simulations where the walls, tables, and chairs around you are all part of the challenge. When the real world and the virtual one interact seamlessly, the possibilities expand exponentially, and we’re just getting started.

How GenAI and Simulation Are Transforming Healthcare and Life Sciences

At this year’s International Conference on Learning Representations (ICLR), NVIDIA Research made it clear that generative AI, simulation, and synthetic data are no longer just promising ideas, they are shaping real-world industries today. With more than 70 papers introduced, NVIDIA showcased advancements that touch nearly every corner of AI, but especially stood out in areas critical to the future of healthcare, robotics, life sciences, and intelligent automation.

NVIDIA's work shows how a full-stack approach, combining faster hardware, smarter algorithms, and optimized AI models, is creating tools that not only learn faster but also apply knowledge in flexible, dynamic ways. From designing new proteins that could unlock the next generation of medicine, to building robots that adapt without expensive hands-on training, to compressing large language models for faster and cheaper deployment, the innovation coming out of ICLR is real and immediate.

“ICLR is one of the world’s most impactful AI conferences, where researchers introduce important technical innovations that move every industry forward,” said Bryan Catanzaro, vice president of applied deep learning research at NVIDIA. “The research we’re contributing this year aims to accelerate every level of the computing stack to amplify the impact and utility of AI across industries.”

Across healthcare, life sciences, robotics, and AI infrastructure, NVIDIA’s latest research highlights one clear message, that the future of intelligent systems will be built with simulation, generative AI, and a whole lot of creative engineering.

Let’s take a closer look at some of the most exciting areas where NVIDIA’s research is pushing boundaries, from healthcare and life sciences to robotics, simulation, audio innovation, and AI efficiency.

Audio + Speech Innovation

Fugatto

Fugatto is the world’s most flexible audio generative AI model. It can create or transform any combination of music, voices, or sounds using a mix of text prompts and audio files. It’s like an AI sound artist that can blend words and audio into completely new creations. NVIDIA also introduced upgrades to audio LLMs to help them better understand and process speech.

Robotics + Simulation Breakthroughs

HAMSTER

HAMSTER is focused on helping robots get smarter without expensive real-world training. By using a layered, hierarchical design, robots can learn new skills by practicing with cheap off-the-shelf data, not actual robot hardware, saving tons of time and money. It's a big step toward better, faster robot learning.

SRSA

SRSA is all about helping robots learn new tricks by building on what they already know. Instead of starting from scratch every time, robots using SRSA can predict which old skill best fits a new task. This smart approach boosted success rates by 19% on tasks the robots had never seen before, huge for real-world applications like healthcare robotics and automation.

STORM

STORM can quickly recreate full 3D outdoor scenes, cars driving, trees moving in the wind, using just a few snapshots. It’s fast (only 200 milliseconds!) and incredibly detailed, making it a perfect tool for training self-driving cars or building realistic outdoor simulations for robotics and smart cities.

Efficient AI Training + Multimodal Learning

Hymba

Hymba is a new family of small, efficient language models that blend transformer models and state space models. This hybrid approach means faster processing (3x better throughput) and much less memory use (4x smaller cache), all without losing performance. Hymba is built for high-resolution memory tasks and better common-sense reasoning.

LongVILA

LongVILA is a powerful new system designed to train AI to understand long videos. It cleverly splits up the training process so it can handle massive video data across hundreds of GPUs at once. This breakthrough lets AI models learn from videos with millions of frames, and it achieves best-in-class results on nine major video benchmarks.

LLaMaFlex

LLaMaFlex introduces a clever new way to shrink large AI models into smaller, faster versions without losing accuracy. It creates compressed models that perform just as well, or better, than models made with traditional methods like pruning. This could dramatically lower the cost and complexity of training families of models.

Healthcare + Life Sciences Transformation

Proteina

Proteina is a transformer-based model that designs the "backbones" of proteins, the core structures that hold proteins together. It’s up to 5x more powerful than previous models, and it could speed up breakthroughs in drug discovery, biotechnology, and healthcare by letting scientists simulate new proteins quickly and accurately.

Reality Capture: Bridging the Physical and Digital Worlds

As the digital and physical worlds become more connected, Reality Capture is emerging as a key technology that's reshaping everything from manufacturing and operations to construction, entertainment, and even historic preservation. It’s not just a buzzword, it’s a true game-changer.

Reality Capture is the process of digitally documenting the real world using tools like 3D laser scanning, photogrammetry, and LiDAR. The result? Highly detailed digital replicas, often called point clouds or visual twins, that give architects, engineers, and creators an exact baseline to design from. It’s already making a huge difference by reducing modeling costs, improving planning, preventing costly errors, and even helping preserve historic landmarks like Notre Dame.

Thanks to advances in cloud storage, drones, and smartphone-based scanning, Reality Capture is becoming more accessible than ever. While it still brings challenges like data management and interpretation, the value far outweighs the hurdles. In the end, Reality Capture isn’t just about copying the world, it’s about understanding it better, with richer detail and deeper insights than ever before.

Read more about Reality Capture here.

Find FS Studio at the Upcoming Conferences

• Boston Robotics Summit

• Florida Simulation Summit

• Automate

• AWE

• AES

Thanks for riding along with us. Have a great weekend!

Until next week, The FS Studio Team