Cosmos3
PulseAugur coverage of Cosmos3 — every cluster mentioning Cosmos3 across labs, papers, and developer communities, ranked by signal.
- 2026-07-01 product_launch NVIDIA released the Cosmos3 foundation model platform for Physical AI applications. source
4 day(s) with sentiment data
-
RynnUnify: A Unified Embodied Model for Planning and Robot Control
Researchers have introduced RynnUnify, a 16-billion parameter unified embodied model designed to integrate visual planning, physical reasoning, and robot control. This Mixture-of-Transformers architecture builds upon Co…
-
Amazon SageMaker JumpStart adds over 100 new AI models in 2026
Amazon SageMaker JumpStart has significantly expanded its catalog in 2026, adding over 100 new models from 30 providers by September. This growth is driven by the rapid release of open-source models and customer demand …
-
New AffectWorld benchmark tests AI's understanding of emotion-driven future predictions
Researchers have introduced AffectWorld, a new benchmark designed to evaluate how well AI models can predict future world states influenced by human emotions. The benchmark focuses on three key areas: World-State Consis…
-
New WATCH method improves transferable world-action models from human videos
Researchers have developed a new cross-video pre-training method called WATCH, designed to improve the learning of transferable world-action models. Unlike traditional methods that focus on predicting the next frame wit…
-
Robots learn dexterous sushi manipulation with tactile feedback
Researchers have developed TacSushi, a novel world-action policy that utilizes tactile feedback to improve robotic manipulation of deformable objects like sushi. This system, built on the Cosmos3 model, learns from reco…
-
Tsinghua AIR's Zeva boosts embodied manipulation success to 73%
Researchers at Tsinghua AIR have developed Zeva, a frozen-weight in-context causal memory system. When integrated with Domain Transform, Zeva significantly improved embodied manipulation success rates from 26% to 73% on…
-
Tsinghua AIR unveils Zeva, enabling embodied AI to learn from experience without weight updates
Researchers from Tsinghua AIR and Domain Transformation have introduced Zeva, a novel embodied AI system that learns continuously without updating its model weights. This In-Context Causal Learning (ICCL) approach allow…
-
NVIDIA launches Cosmos3 foundation model for Physical AI
NVIDIA has released its Cosmos3 foundation model platform, designed to accelerate the development of Physical AI applications. The platform includes several omnimodal world models capable of generating video, image, aud…
-
New benchmarks and models advance vision-language capabilities in robotics and reasoning · 10 sources tracked
Recent research explores advancements in vision-language models (VLMs) across several domains. DeCAL introduces a new model for dexterous manipulation that integrates tactile sensing and visual-language understanding. R…
-
ACE ROBOTICS' Kairos World Model Sets New Embodied AI Benchmark
ACE ROBOTICS has released its Kairos world model, which has achieved top rankings in four major embodied AI benchmarks: RoboTwin 2.0, LIBERO-Plus, WorldModelBench Robot, and DreamGen. The model utilizes a novel unified …
-
Embodied Intelligence Adopts Bird's-Eye View for Scalable Robot Data
A new approach called Dexterity-BEV is being introduced to address the data challenges in embodied intelligence by adapting the Bird's-Eye View (BEV) methodology from autonomous driving. This method aims to unify hetero…