I'm going to tell you something.The shadows
A robotics forecaster reported on November 5 that just now, Beijing's Physical World Model-creating startup, PCT, had announced on X that it would be launching a new generation of robots.Official launch of the TensorWorld brand。
It's the first time I've ever heard of it.
At the same time, the company disclosedA round of seed funding of US$105 million (about US$705 million) has been completed.It was founded in July 2026 and is dedicated to building cutting-edge models that can understand, simulate and interact with the real physical world.Basic models of physicsThe aim is to enable the model to understand the structure and operation of the physical world, to understand the real world, to predict the outcomes of different actions and to provide the basis for subsequent decisions. The goal is to enable the model to apply the learned physical knowledge to different tasks and environments, even in the face of unfamiliar scenarios, to reason and predict, and to continue learning based on new information.Physical simulation, timing modeling, three-dimensional understanding and agent planningThere are a number of research results in various fields.He is the founder and CEO of IoT Technology.He has published seven papers in top academic conferences and journals such as ICML, NeurIPS and others, six of which were first authored or co-authored first. He then went to MIT CSAIL as a postdoctoral researcher with Professor Wojciech Matusik, a renowned scholar in the field of computational design and manufacturing.
He is also the co-founder and CEO of the company.
His research focuses primarily on deep learning and scientific machine learning, particularly the modeling of physical systems and dynamical systems.Neural network physics solver Transolver, pre-trained model GeoPT, and two models of TimesNet and AutoformerIn this study, Transolver focuses on the study of complex operating laws, simulating and predicting changes in physical systems under different conditions. GeoPT allows models to learn the relationship between geometric structures and physical dynamics through large-scale data pre-training. The study also won the best paper award at the ICLR 2026 Science foundation model Workshop. TimesNet captures the changing characteristics of different cycles in the data. Autoformer is used to analyze the changing trends and cyclical laws of time sequences to predict future data changes.What's going on?Participation in research and developmentUni3D modelIt is primarily used to study the shape, structure, and other characteristics of three-dimensional objects in order to improve the understanding of three-dimensional objects.Weirui YeIt has been suggestedEfficient Zero strengthening learning algorithmIt enables Agents to make use of limited interactive experience to learn and to plan follow-up actions using the learned world models.
Wen ZhengParticipation in the creationMore than 100 million usersAs part of the build-up to a computing infrastructure that supports more than 100,000 accelerators, the company says it will use this experience to build large-scale cutting-edge models and promote the transformation of scientific results into real products.
