Independent global robotics news from Article module 102.
Each record is classified by both destination market and product layer, including complete robots and core components.
A University of Hong Kong team presents the YN decision model, reporting 37,000 tokens/s and million-scale household simulation. The article explains the model’s decision...
Researchers reproduce BeyondMimic on the Unitree G1 humanoid robot and train one reinforcement-learning policy to perform 31 motions. The article covers the data, simulat...
Simulation can accelerate robot learning, but it cannot reproduce every contact, material property, failure mode, or deployment constraint. This article explains why real...
Daimon Robotics is building tactile intelligence around both data collection and model design. The article explains how tactile sensors, datasets, and world models can be...
Human beings need to represent the shape and direction of an object in their minds, understand the relationship between different points of view, and infer three-dimensio...
Human-powered robots are moving from a "power-driven" demo to a "power-driven" commercial phase, where the real mass-production rhythm is stuck, often not with large mode...
A graduate student’s observations from Silicon Valley offer a candid look at building an embodied-AI company, including the gap between research prototypes, hardware cons...
The first author of the paper, Zhang Yu, of the South China University of Technology MMLab, was a PhD student, studying the direction of physical simulation and the combi...
From GPT-6 Astra, Claude Opus 5.5 to DeepSeek V4.1-Flash-1017, Kimi K3.1 Pro, three-dimensional creation is becoming a new scene for demonstrating model capabilities: bui...
It sounds natural, but for robots, this is precisely a long-standing problem. For example, allowing robots to pick up cups from the table. Going to the table does not mea...
With the improvement in model capabilities, model evaluation also faces new challenges. Part of the evaluation benchmarks are close to zero, and the model score is increa...
This is one of the core shortcuts of human robots today. We humans stretch out our hands and take things without having to stare at them, finger-touching will tell us how...
Imagine a scenario like this: a natural language instruction is given to a robot, and it passes through the corridor, converges, and eventually enters a designated room. ...
In the past few years, robotic operating models have gradually moved from single-action to multi-tasking and time-control. Models have begun to process the changes in lan...
01 Full-focused vision matrix, deep-farming semiconductor core scene R5000Pro series AI intelligent reading codes for multi-height, variable-distance industrial reading s...
One person picks up a cup on a table and the camera can record the entire process of action. But how to extract the camera's trajectory, hand movements and changes in the...
With the accelerated integration of AI models with robotic control, how to make robots more convenient to access end-to-end devices, complete crawling tasks, and become o...
SPADER: Step-wise Peer Advantage with Diversity-Aware Exploration Rewards for Multi-Answer Question Answering Entry Information: EMNLP 2026 Main Conference RL method Ther...
A recent study shows that GPT-6 Astra has solved 98% of all new theoretical physics problems! This top-level theoretical physics review, TheoPhysEval, has just shocked th...
It has no propellers, no springs, no control rotor, no mechanical device specifically designed to maintain balance. The entire robot has only five driving freedoms, less ...