A single destination for timely, editor-curated robotics news from around the world.
WorldOracle, a new benchmark for action-conditioned world models (ACWMs), has been introduced by researchers from Princeton University, the University of Texas at Austin, the University of Toronto, NVIDIA, and AgiBot. This platform evaluates the predictions of various models against actual manipulat...
WorldOracle has introduced a benchmark to assess action-conditioned world models (ACWMs) by comparing their predictions with actual robot executions. In a demonstration, a robot's grasp failed while the model inaccurately predicted success, highlighting discrepancies between simulated and real-world...
Anyverse Dynamics has launched and open-sourced MWA-Phoenix, a 7 billion parameter latent world action model designed for robotic applications. This model utilizes only RGB input and has achieved top scores on six different robot benchmarks, including performance in real-robot testing. The introduc...
NVIDIA has enhanced the Cosmos platform to assist developers in building world models that enable physical AI systems to comprehend and simulate the real world. The updates include new training tools, simulation checkpoints, and Cosmos Data Factory blueprints, which rank among the top five on six op...
The 2026 Physics AI and World Model Conference is set to be held in Nanjing, marking a significant event as the first high-profile industry conference in China dedicated to this field. This conference aims to gather experts and innovators to discuss advancements in physics and artificial intelligenc...
Skyfall AI has introduced SKY-001, a world model aimed at enhancing the design of complex physical products through advanced spatial reasoning and planning. This innovative model is reported to be 100 times faster and operates at 150 times lower compute costs compared to existing models, with a comm...
WorldOracle has introduced its World Model Evaluation Platform, designed to conduct real machine testing to evaluate robotic systems. This platform aims to address the industry's need for reliable assessments of robotic capabilities, particularly in the context of ethical and regulatory challenges f...
The ongoing discourse around the concepts of 'Generation' and 'Scanning' in world model development is gaining traction. This division highlights the contrasting approaches to artificial intelligence and its applications in various sectors, including manufacturing and home environments. Understandi...
UBTECH has announced that its Thinker-WM has achieved top scores in three embodied world model benchmarks. The integration of depth data significantly improved single-task success rates on the RoboTwin platform, increasing performance by 15 percentage points during ablation tests. This advancement...
Hesai has introduced Kosmo, a 3D world data platform that integrates the Kosmo One spatial camera and Kosmo World cloud service to transform real-world environments into digital formats for robot training and world modeling. An operator can scan a typical 100-square-meter indoor area in approximatel...
Hello Inc.'s robotaxi division has introduced HelloWorld, a generative driving world model designed to create multi-view camera video and lidar data. This innovative model can generate these outputs based on edited layouts or plain-language prompts, enhancing testing capabilities for autonomous vehi...
AMD has signed a definitive agreement to acquire World Labs in an all-stock transaction valued at approximately $8.2 billion, expected to close by the end of 2026. This acquisition marks a significant investment in the 4D world modeling sector, with World Labs founder Fei-Fei Li joining AMD as Execu...
DexForce has introduced UNITAS, the world's first native 3D world model designed specifically for robotic tasks. This innovative model aims to unify the understanding of physical interactions by allowing robots to comprehend the three-dimensional physical world, which is crucial for advancing toward...
Odyssey, an AI research company, has launched Odyssey-3, its most advanced world model to date. This model represents a major leap towards general-purpose intelligence, enabling systems to understand, predict, and interact with both physical and virtual environments. Odyssey-3 can control robots, hu...
Moxin Technology has introduced its MoWorld application, which transforms pet photos into animated 4D digital twins. This innovative feature is available on Huawei's Mate 90 smartphone, utilizing Ascend cloud generation technology alongside a sophisticated 600M-parameter model powered by Kirin. The...
AIsphere has launched PixVerse R2, a groundbreaking real-time world model capable of generating explorable audiovisual environments. This innovative platform takes various inputs, including text, images, audio, and keyboard commands, while maintaining state throughout a session. The introduction of...
Odyssey has unveiled Odyssey-3, a versatile world model designed to control robots, humanoids, autonomous vehicles, and drones using a unified AI system. This autoregressive diffusion transformer is trained on a diverse set of visual observations, allowing it to adapt with minimal task-specific data...
On September 14, Shanghai Juna Technology Co., Ltd. officially launched Cog-WM 1.0, the world's first brain-inspired cognitive world model, at the 2026 Pujiang Innovation Forum. This model, based on systematic brain-like neural mechanisms, has demonstrated capabilities in autonomous navigation and m...
Shanghai Artificial Intelligence Laboratory has introduced the Intern physical world model W0, which features native force-tactile sensing and duplex collaboration capabilities. This model is designed to enhance robotics applications by integrating with Intern InkStone and the science model S2, faci...
On September 14, Digua Robotics and Giga Vision announced a strategic collaboration focused on integrating their respective technologies. Giga Vision will provide world models, embodied foundational models, and real-world application experience, while Digua Robotics will contribute an edge AI comput...
On September 10, at the Bund Conference, Luo Yihang, co-founder and CEO of Shengshu Technology, unveiled Motus2, a self-evolving general world model aimed at robotic dexterous manipulation. Unlike traditional robots that rely on explicit instructions, Motus2 enables robots to learn autonomously from...
ACE Robotics and NTU S-Lab have announced the open-source release of Puffin-World, a multimodal world model that integrates physics, geometry, and appearance. This model utilizes the Puffin-16M dataset and achieves state-of-the-art camera absolute pose errors, reporting sub-degree accuracy across fo...
As artificial intelligence (AI) evolves from digital interactions to real-world applications, it faces complex challenges in perception, decision-making, and action. World models are seen as a promising solution, enabling AI to predict outcomes and adapt to environmental changes. This technology is...
AgiBot has introduced GE-Act 2.0, a native world-action model that has been pretrained from random initialization using embodied data. This new model has significantly scaled from 300 to 30,000 hours of training, enabling it to perform zero-shot skills, including towel folding, across two different...
Scalabot has introduced the HERON-World Model, a new action-conditioned world model aimed at enhancing robotic understanding and future prediction capabilities. This model utilizes a three-tier data pyramid, incorporating real robot teleoperation data, simulation data, and first-person human operati...
World Labs has unveiled Atlas, a new world model aimed at generating, reconstructing, and simulating 3D environments. This model supports Real-to-Sim workflows, allowing robots to navigate and manipulate spaces reconstructed from images or videos. Atlas can create RGB and depth data that robots woul...
Florent Delgrange received the Best Blue Sky Paper Award at AAMAS 2026 for his research on Foundation World Models for agents that adapt in dynamic environments. His work addresses the challenge of ensuring that autonomous agents continue to learn effectively while maintaining reliable behavior in c...
From August 19 to 23, 2026, the World Robot Conference (WRC) took place at the Yichuang International Exhibition Center in Beijing. The event, themed 'Human-Machine Symbiosis, Integration of Production and Demand,' featured over 300 exhibitors and more than 2,000 exhibits. Keenon Robotics showcased...
Samsung Electronics is accelerating the development of its proprietary humanoid robot by integrating internal manufacturing data, custom hardware, and foundation AI models into a centralized initiative. Reports indicate that this effort operates independently from Rainbow Robotics, where Samsung hol...
Beta Infinite has introduced its BetaWAM 0.1 world action model, showcasing a robot capable of autonomously completing complex, long-range tasks in dynamic environments. This model addresses the limitations of existing technologies that primarily focus on single-point actions or fixed operations. T...
Li Yin, the former CTO of Huawei's Pangu model, has launched a startup named Xirang Kaiwu Technology, focusing on world models. The company, registered on July 1, 2026, in Shenzhen, has an initial capital of 1 million yuan and a valuation of 1.5 billion yuan following its recent funding round. Li ho...
LTX has introduced LTX-2.5, an advanced version of its open-weights world model, enhancing capabilities for video generation and physical AI. This model boasts improvements in visual quality, prompt understanding, and generation speed, allowing developers to customize it on their hardware. With over...
Sota Unbounded has partnered with Europe's largest supermarket group to deploy over 1,000 embodied intelligence robots in real retail environments over the next three years. This strategic collaboration aims to break the commercial stagnation in the field of embodied intelligence, which has seen sig...
On August 10, 2026, Daimon Robotics unveiled the world's first Tactile-grounded World Model (Daimon-TWM), showcasing a robot's ability to adapt to unexpected disturbances without stopping or colliding. This demonstration challenges the conventional belief that robots can only execute standardized pr...
A startup named 'Xindu Qiyuan' has successfully utilized a 'white box world model' based on 'uncertain differential geometry' to enable a humanoid robot to perform a complete task from voice command to cup retrieval without prior calibration or pre-training. This achievement is particularly signific...
NVIDIA has joined over 200 organizations in signing an open letter advocating for AI leadership through open ecosystems. This approach emphasizes the importance of open models in physical AI, which require understanding and predicting environmental behaviors rather than just appearances. The signif...
Kuwah Technology has appointed Dr. Yang Xue as Chief Scientist to enhance its world model productivity. The company is also collaborating with Shanghai Jiao Tong University to establish a joint laboratory focused on embodied intelligence and world models. These initiatives aim to improve the efficie...
Muka Robotics has achieved a remarkable feat with its LJM Latent Joint-conditional Model, securing the second position globally at WorldArena with a motion quality score of 89.17 using only 32 Zhenwu 810E GPUs. This accomplishment highlights the model's innovative dual-expert architecture, which emp...
At the WAIC 2026 event, Muka Robotics, Shengshu Technology, EvoPhys.ai, and Chengwei Capital highlighted a significant shift in the global tech landscape. They noted that Chinese world model startups are now operating independently, without American counterparts to benchmark against, which challenge...
The BYD AI team has introduced the HyWorldVLA, a hybrid pixel-latent world model that utilizes VLA architecture. This model has achieved a score of 90.59 PDMS on the NAVSIM benchmark, marking a significant milestone for BYD in the field of autonomous driving foundation models. This achievement is n...
Contactile has developed tactile sensors for robotic hands and grippers that enhance world models in robotics. These sensors address the limitations of current conditioning methods, which primarily rely on visual inputs and joint encoders, by incorporating friction as a critical input. This advancem...
Moonshot AI has introduced Kimi K3, a groundbreaking 2.8 trillion-parameter open-source AI model, marking it as the largest of its kind to date. This model is designed to handle complex workflows and features a one-million-token context window, enabling it to perform tasks significantly faster than...
On July 17, Extreme Intelligence unveiled Gravity 4D WAM, the first module of its embodied intelligence framework, at the 2026 World Artificial Intelligence Conference (WAIC). This launch addresses a critical issue in current visual language models: the discrepancy between visually plausible predict...
The World Artificial Intelligence Conference (WAIC) 2026 highlighted the significance of world models as a crucial pathway for physical AI to transition from laboratory settings to real-world applications. The event featured a summit focused on the integration of world models and embodied intelligen...
PHANES AI, established by 28-year-old Yang Shuo from HIT, has introduced TouchWorld, a tactile foundation model designed to enhance robotic manipulation capabilities. This model enables robots to perform precise physical tasks by incorporating a sense of touch, marking a significant advancement in r...
Artificial intelligence is evolving with a focus on 'world models,' which simulate environmental responses to actions. This shift is gaining traction among Chinese companies, expanding the application of these models beyond traditional physics and robotics. The technology is still developing, with n...
1X Technologies has announced the establishment of the 1X World Model Lab, a new research initiative focused on creating foundational AI models from the ground up. This strategic shift comes as the company faces significant leadership turnover, including the recent hiring of Samarth Sinha from Luma...
At COMPUTEX in Taipei, NVIDIA unveiled Cosmos 3, a groundbreaking open world foundation model designed to integrate vision reasoning, physical simulation, and action prediction. This launch represents a significant shift in the robotics industry, moving from language-centric frameworks to video-firs...
In the spring of 2026, the term 'world model' has become central to discussions surrounding robotic foundation models, as noted by researcher Chris Paxton. The term's growing popularity has led to confusion, as it carries different meanings across various AI labs, each with unique strengths and weak...
AGIBOT has introduced the GO-3 Preview, an embodied AI model that utilizes a World Reasoning Action Model (WRAM) architecture. This innovative model connects reasoning, prediction, and action generation, enabling robots to learn from human videos, simulations, and real-world experiences. The GO-3 Pr...
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
Daily robotics news, in-depth analysis, conference highlights, and discussions with professionals worldwide.
