X-WBC: A Cross-Embodiment Foundation Model for Humanoid Whole-Body Control
X-WBC is a cross-embodiment humanoid whole-body control foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. It uses human-centric command tokens to align full-body human motion, robot reference motion, and sparse VR observations; a causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, and lightweight robot-specific modules then map the shared representation to each robot's proprioception and action space.
Highlights
AgiBot Releases Genie Studio 2.0 Embodied Intelligence Development and Deployment Platform
AgiBot released Genie Studio 2.0, an embodied intelligence application platform covering development, deployment, and operations, composed of three modules—Flywheel, Agent, and Monitor—that connect data collection, model training, evaluation, inference, task orchestration, and robot fleet management.
Tsinghua AIR Proposes OxyGen: Unified KV Cache Management Lets Robots Talk While Acting
The Institute for AI Industry Research (AIR) at Tsinghua University, together with the Department of Electronic Engineering at Tsinghua University and the University of Science and Technology of China, proposed OxyGen, which through unified KV Cache management lets an embodied model's action and language reasoning share prefix computation; the paper has been accepted by NeurIPS 2026.
CVer · 微信公众号Source ↗RynnVLA-Latent: Scaling Robot Manipulation Foundation Models with 3D-Grounded Latent Actions
The research team proposed the RynnVLA-Latent framework, which through RynnLAM learns continuous, 3D-grounded latent actions from visual changes without requiring native robot control signals, converting 30.1K hours of human, robot, and simulation video into 745.52M latent action descriptions for policy pretraining.
arXiv cs.ROSource ↗
Model Release/Update
Kinetix AI Releases KAI World Model, Supporting Locomotion and Manipulation Tasks
Kinetix AI released the KAI World Model, which is based on parallel video generation and 4D spatial modeling and can simultaneously generate a virtual world and reconstruct spatial scenes. The model is pretrained on millions of videos, supports pure locomotion, pure manipulation, and coordinated locomotion-and-manipulation tasks, and claims generalization across tasks and across robot embodiments. Real-robot demonstrations include a robotic arm autonomously opening a beverage can and a humanoid robot playing real-time table tennis rallies.
Product Release/Update
AgiBot Releases Genie Studio 2.0 Embodied Intelligence Development and Deployment Platform
AgiBot released Genie Studio 2.0, an embodied intelligence application platform covering development, deployment, and operations, composed of three modules—Flywheel, Agent, and Monitor—that connect data collection, model training, evaluation, inference, task orchestration, and robot fleet management.
Industry News
USV and Creandum Co-Lead Machenta's $40 Million Seed Round
USV announced that it co-led Machenta's $40 million seed round with Creandum, and the company simultaneously emerged from stealth. Machenta claims its robot foundation model is trained entirely in simulation without real-world data, can learn a task from a single demonstration without additional training, and can adapt when the environment changes.
Union Square Ventures · XSource ↗
Paper Research
X-WBC: A Cross-Embodiment Foundation Model for Humanoid Whole-Body Control
X-WBC is a cross-embodiment humanoid whole-body control foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. It uses human-centric command tokens to align full-body human motion, robot reference motion, and sparse VR observations; a causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, and lightweight robot-specific modules then map the shared representation to each robot's proprioception and action space.
arXiv cs.ROSource ↗Tsinghua AIR Proposes OxyGen: Unified KV Cache Management Lets Robots Talk While Acting
The Institute for AI Industry Research (AIR) at Tsinghua University, together with the Department of Electronic Engineering at Tsinghua University and the University of Science and Technology of China, proposed OxyGen, which through unified KV Cache management lets an embodied model's action and language reasoning share prefix computation; the paper has been accepted by NeurIPS 2026.
CVer · 微信公众号Source ↗RynnVLA-Latent: Scaling Robot Manipulation Foundation Models with 3D-Grounded Latent Actions
The research team proposed the RynnVLA-Latent framework, which through RynnLAM learns continuous, 3D-grounded latent actions from visual changes without requiring native robot control signals, converting 30.1K hours of human, robot, and simulation video into 745.52M latent action descriptions for policy pretraining.
arXiv cs.ROSource ↗Ropedia Releases Six Research Results, Exploring How First-Person Human Data Can Be Transformed into Robot Capabilities
Ropedia released six studies in a concentrated manner during the first Ropedia Super Week from October 19 to 23, centered on how first-person human video can be transformed into training and execution information usable by robots.
机器之心 · 微信公众号Source ↗
Brief updates
AgiBot Team Releases GE-Act 2.0 World-Action Model Technical Report
The AgiBot Research Team released the GE-Act 2.0 world-action model technical report, in which the generation and action components are all trained from scratch on manipulation data, consisting of the control-oriented autoencoder CoAE, the single-step visual planner SVP, and the inverse dynamics model IDM, and jointly trained with KASO.
arXiv cs.ROSource ↗