World Models

AIAI-generated summary Google Research introduces an investigational machine learning framework that translates passive wearable signals into a modifiable coachable health age to track early cardiovascular-kidney-metabolic health. Google Research says the framework uses an action-conditioned world model to simulate personalized 12-week what-if habit trajectories and expands to routine blood biomarkers.

Google World Models AI for Science Healthcare

AIAI-generated summary Researchers at MIT's Computer Science and Artificial Intelligence Laboratory and Carnegie Mellon University developed a method called Workspace Models to help robots plan actions rapidly. Nitish Dashora says the system offers a latency upgrade over agentic approaches by training a smaller module to memorize key details.

MIT World Models Robotics

AIAI-generated summary NTT, Inc. announced a physics-based world model designed to evaluate robot actions prior to execution by predicting outcomes such as drops and collisions. NTT says the technology increased task success in a small-robot shelf-loading test from 73.1% to 92.3% and plans field trials with Mujin, Inc.

World Models Robotics

AIAI-generated summary The Institute of Foundation Models at MBZUAI announced the release of PAN 2, providing researchers with an open video simulator that includes model weights along with training and inference code. The model uses a 29 billion parameter diffusion model alongside a Qwen3.5-9B text encoder and a Wan 2.2 VAE to generate video clips up to 20 seconds long.

World Models Video Generation Open-Weight Models

AIAI-generated summary Google DeepMind reports that Gemini 4 Argon robotics research capabilities acted as a simulation engine to accelerate the development of Gemini Robotics 2.5. Google DeepMind says Argon generates simulated practice runs from camera scenes and builds digital twins to help robots generalize to new objects.

Gemini Google DeepMind World Models Robotics VLA Models

AIAI-generated summary Machine learning researchers have developed a technique to help physical AI systems and robots make zero-shot physics inferences using counterfactual world models. Randall Balestriero of Brown University and AMI Labs, Kerda Varaku, and Yann LeCun describe the approach in a pre-print paper titled Counterfactual World Models: Natura Non Facit Saltus.

Yann LeCun World Models Robotics Physical AI

AIAI-generated summary Researchers at MIT, the University of British Columbia, and other institutions developed a single deep-learning system named WorldParticle to simulate physical processes like sliding sand and folding proteins. The system represents simulated materials as tiny particles and predicts their movement step by step, allowing engineers to model multiple physical phenomena with one unified architecture.

MIT World Models AI for Science

AIAI-generated summary Machine learning researchers have developed a new technique designed to help world models make better predictions about how physical objects will behave. Randall Balestriero, Kerda Varaku, and Yann LeCun describe this approach in a pre-print paper titled Counterfactual World Models: Natura Non Facit Saltus.

Yann LeCun World Models

AIAI-generated summary The Conference on Robot Learning features keynotes comparing Fei-Fei Li's world models from World Labs with Russ Tedrake's factory-trained behavior models from Walden Robotics. World Labs describes Atlas as a multimodal autoregressive diffusion transformer that generates 3D scenes, while Walden Robotics reports its robots have worked in a Toyota plant.

Fei-Fei Li World Models Robotics VLA Models

AIAI-generated summary Niantic Spatial chief technology officer Brian McClendon says the company is focused on reconstructing the real world using visual inputs instead of text. Odyssey CEO Oliver Cameron says world models predict the next visual state of the world to generate simulations that look and feel real.

World Models Spatial Intelligence

AIAI-generated summary Alex de Vigan launched Physicl after his previous business shifted focus due to the rise of generative AI. The startup secured $10 million in seed funding from Inovia Capital to help robot builders and world model designers train models using realistic simulations.

World Models Robotics Startups Funding

AIAI-generated summary AGIBOT unveiled its GO-3 Preview robot model on Oct. 29 at its Innovation Day in Shanghai, featuring a system built on a world reasoning action model. AGIBOT says the model ranks first on three benchmarks, though the company supplied those rankings itself and has not yet released the weights or code to the public.

World Models Robotics VLA Models

AIAI-generated summary Chatbots powered by large language models like ChatGPT, Claude, Google Gemini and Perplexity AI can write poems, brainstorm and solve complex problems. World models are an emerging technology that could help artificial intelligence systems anticipate events and use predictions to act.

Google Perplexity Claude ChatGPT Gemini

AIAI-generated summary Patronus AI introduces GlobeBench, a new benchmark featuring 1,632 tasks designed to evaluate text-based world models across five domains and 19 simulation skills. GPT-6 Astra achieved the highest success rate at 49.57%, while Patronus AI notes that no model solved half of the benchmark tasks.

World Models Benchmarks

AIAI-generated summary NVIDIA released Cosmos, a platform of world foundation models and development tools designed to address physical AI challenges. NVIDIA says Cosmos Framework provides agent skills and recipes for training, while Cosmos NIM microservices support GPU-accelerated serving.

NVIDIA Foundation Models Model Releases World Models

AIAI-generated summary Flexcompute has released its new Physics Twin for Space Plume Impingement simulation tool to help mission teams with complex rendezvous and proximity operations. According to the company, the software evaluates thruster exhaust risks on nearby spacecraft and reduces computational processing time compared to hybrid Direct Simulation Monte Carlo setups.

World Models

AIAI-generated summary Baidu unveils Apollo World, a world model built to train autonomous vehicles such as Baidu robotaxis at its AI Day Apollo event. Baidu says the model can generate about three years of real-world testing per day after being trained on 217.48 million miles of data from the Apollo Go fleet.

World Models Autonomous Driving Physical AI

AIAI-generated summary LiberAI announced Liber-0, a world action model pretrained on more than 2.2 million hours of human experience to support zero-shot robot manipulation. LiberAI says the model achieved an overall success rate of 36.0% across 60 tasks evaluated in 600 trials without robot demonstrations collected specifically for those target tasks.

World Models Pretraining Robotics VLA Models

AIAI-generated summary Israeli startup 42AI has launched with $62.5 million in seed funding and a behavioral world model designed to predict human behavior. The company is initially running its predictive system at a handful of financial brokerages to help predict how individual traders will respond to volatile conditions.

World Models Startups Funding

AIAI-generated summary According to a recent paper, separately trained artificial intelligence systems reportedly find a common base of communication under clear conditions. This mathematical proof provides a recipe for achieving the translation process using significantly less data.

World Models Translation