AI Tech Flash ── 2026-10-24
Nvidia adds tools to generate rare robot training scenarios
Nvidia expands Cosmos with tools for synthetic data that is hard to collect in reality
What changes
Developers of robots and autonomous vehicles can prepare synthetic footage of situations that are hard to collect in the real world. According to MIXED Reality News, these include rare events, manufacturing defects and changes in weather or lighting.
They also gain more ways to test policies (the rules that decide a robot's actions) in simulation before deployment. However, MIXED Reality News notes that the ANYbotics example is not an independently measured improvement in robot safety.
Why it matters
According to MIXED Reality News, Nvidia describes the tools as a way to generate training examples that would be difficult or unsafe to collect in the real world. This targets a common data shortage in industrial robot and autonomous vehicle development.
The outlet says the tools are aimed at developers rather than a consumer video-generation app. The main addition is software for building and testing physical-world AI before a robot or vehicle faces the same conditions outside simulation.
What it is
Synthetic training scenario generation means using AI to create scenes that are hard to film in reality and using them to train other AI systems. Nvidia's Cosmos is a set of tools for this.
According to MIXED Reality News, Nvidia expanded Cosmos in an October 22 announcement. It covers data preparation, model training and simulation. It can produce synthetic video of rare events, manufacturing defect images, and changes in weather or lighting.
Who and where
Nvidia builds it. According to MIXED Reality News, it is aimed at developers building systems such as industrial robots and autonomous vehicles.
The platform is available through GitHub and cloud services including Microsoft Azure, Nebius and CoreWeave. Developers can use individual components rather than adopting every part of the Cosmos stack.
Why it was built
Robot and vehicle AI needs training data from many kinds of situations. According to Nvidia's announcement as reported by MIXED Reality News, some of those examples are difficult or unsafe to collect in the real world.
The new tools were built to fill that gap with synthetic examples. MIXED Reality News describes the aim as letting developers build and test physical-world AI before exposing a robot or vehicle to real conditions.
How it works
Cosmos Data Factory now offers five workflows: generating autonomous-driving datasets, varying the appearance of objects in images, turning still images into event videos, creating defect images, and altering weather or lighting in existing footage. The last one preserves the scene structure.
For simulation, new Cosmos 3 Nano variants support scenarios with multiple camera views and sensor inputs. Accompanying code lets developers adapt them to evaluate autonomous vehicles and two-armed robots in interactive simulations.
The Cosmos Framework also extends fine-tuning, reinforcement learning, distillation and quantisation to compatible open models. Distillation and quantisation are methods for making models smaller and lighter.
Key points
- On October 22, Nvidia announced new Cosmos tools for synthetic data on rare events, defects and changing conditions.
- Nvidia says ANYbotics generated more than 1,500 synthetic clips; this is a reported example, not an independent measurement.
- Nvidia reports up to twice the inference performance from updated Cosmos 3 NIM services in validated configurations, not across all models and hardware.
Conclusion
Nvidia has expanded Cosmos so developers can synthesize training scenes that are hard to collect in the real world. It is available through GitHub and several clouds. The ANYbotics example and the performance figure are Nvidia's own reports, and no independent verification is cited.
Quotes and sources
“Nvidia has expanded Cosmos with tools for generating training examples that would be difficult or unsafe to collect in the real world.”── MIXED Reality News, 2026-10-24
“Nvidia says inspection-robot company ANYbotics generated more than 1,500 synthetic clips to help train systems to recognise liquid spills and other changes between site visits.”── MIXED Reality News, 2026-10-24
“The platform is available through GitHub and cloud services including Microsoft Azure, Nebius and CoreWeave.”── MIXED Reality News, 2026-10-24
“Nvidia reports up to twice the inference performance from updated Cosmos 3 NIM services in validated configurations.”── MIXED Reality News, 2026-10-24
Source articles
- Hydrocarbon Engineering 10/23 08:03 UTC
- MIXED Reality News 10/24 08:39
Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.