The article reports on major capital events in the recent robotics and embodied intelligence sectors, including AMD's acquisition of WorldLabs, a company co-founded by Fei-Fei Li, for approximately 55 billion yuan, Midea Group's plan to acquire a welding subsidiary for 6.64 billion yuan, and several robotics companies such as Benmo Technology, Huanchuang Technology, and Juxi Intelligent completing financing or going public. Additionally, multiple embodied intelligence products and solutions have been launched, covering humanoid robots, service robots, flexible electronic skin, tactile perception systems, and more, showcasing the industry's rapid progress in technology, capital, and application scenarios.
Skild AI introduces their omni-bodied robotics foundation model, the Skild Brain, which is trained across diverse morphologies and data sources to enable generalization across tasks and hardware. The model uses a hierarchical architecture with low- and high-frequency policies, and is pre-trained using large-scale simulation and internet video data, with post-training on real-world data. This model represents a significant step toward creating a general-purpose robotic brain capable of operating on various robot types.
Editorial context：Skild AI introduces their omni-bodied robotics foundation model, the Skild Brain, which is trained across diverse morphologies and data sources to enable generalization across tasks and hardware. The model uses a hierarchical architecture with low- and high-frequency policies, and is pre-trained using large-scale simulation and internet video data, with post-training on real-world data. This model
Skild AI presents a research paper detailing the development of an AI model trained across a vast array of robot bodies, enabling it to adapt to unpredictable scenarios without prior exposure. The model demonstrates zero-shot control and in-context learning, showing resilience in scenarios like limb loss, joint failure, and morphological changes. The work highlights the importance of adaptability in embodied AI for real-world applications.
Editorial context：This research introduces an 'omni-bodied' AI model trained across 100,000 different robot bodies, demonstrating zero-shot adaptation to extreme morphological changes through in-context learning. The approach emphasizes the need for AI to adapt rather than memorize, drawing parallels to biological evolution and AGI development.
To address the issue of insufficient warehouse space, Figure AI decided to melt down the F.02 robot instead of disassembling or selling it. This decision involves multiple considerations, including intellectual property protection, technical confidentiality, and asset management. Although the destruction process faced challenges in terms of site and technology, it was ultimately completed at the Finnish factory. This incident has also sparked discussions about the management of retired robots and industry standards.
Editorial context：The destruction of Figure AI's F.02 robot reflects the complexity of robot decommissioning management, involving multiple considerations such as intellectual property protection, asset management, technical confidentiality, and industry standards. The choice of destruction method is not only related to technical implementation but also reflects the company's emphasis on data security and commercial confidentiality.
2026-09-29Tue
Tuesday
Robotic Systems Lab@leggedroboticsSignalEditorial score6262
Editorial context：The work introduces LP-ACRL, a curriculum learning framework that automatically samples terrain types, difficulty levels, and velocity commands based on policy performance, eliminating the need for pre-defined training sequences.
The article explores the issue of data scarcity in the field of embodied intelligence, pointing out that although there are already a large number of data collection centers, high-quality and reusable data remains severely insufficient. The article analyzes structural contradictions such as high data collection costs, inconsistent quality, fragmented formats, and lack of cross-ontology reusability. It introduces Wu Wen Tech's solution of building a data foundation through a Real2Sim2Real closed-loop system, including large-scale data collection, automated annotation, and simulation training technologies, ultimately forming a data-driven flywheel to promote the development of embodied intelligence.
Editorial context：The article highlights the critical data scarcity in embodied AI, emphasizing the gap between the scale of robot models and the availability of high-quality physical interaction data. It identifies structural issues such as high collection costs, quality inconsistencies, and lack of standardization as major barriers. The solution proposed by Wu Wen Tech involves a Real2Sim2Real closed-loop system,
Skild AI announces a breakthrough in physical AI through self-play, where a model like S1 learns to perform complex tasks such as soccer by competing against itself in simulation. The approach demonstrates the potential for robots to exceed human capabilities, drawing inspiration from historical AI advancements in game-playing.
Editorial context：Skild AI introduces a new approach to physical AI through self-play, where a model like S1 learns complex tasks by competing against itself in simulation. This method shows promise for surpassing human capabilities in robotics, drawing parallels to historical successes in game-playing AI like AlphaGo and AlphaStar.
This article explores how tactile data is helping robots improve their dexterous manipulation skills. Researchers are creating large tactile datasets and models that can use tactile feedback to enhance robot performance in tasks like folding laundry or turning keys. Challenges include the difficulty of integrating tactile data with vision-based models and the need for more diverse and scalable datasets to achieve significant improvements in robot dexterity.
Editorial context：The article highlights the growing importance of tactile data in advancing robot dexterity, emphasizing the challenges of integrating tactile feedback into vision-language-action (VLA) models. It discusses recent research efforts to create diverse tactile datasets and models that can generalize across different robotic hardware, while also addressing the limitations of current approaches and the '
Researchers from MIT Lincoln Laboratory deployed sensors in the Arctic to monitor under-ice sounds and test through-ice communication using a magnetic modem. The study addresses challenges in extreme weather and highlights the importance of community collaboration and sensor deployment strategies for Arctic research.
Editorial context：This research focuses on through-ice communication and acoustic monitoring in the Arctic, using low-cost sensors and magnetic modem technology. The study highlights the challenges of deploying and retrieving equipment in extreme conditions and emphasizes the importance of community engagement and collaboration with local experts.
FedEx and Dexterity have expanded their collaboration to deploy Dexterity's Foresight world model and Mech trailer loading systems at the FedEx Hagerstown Hub. This marks a significant scaling of physical AI in logistics, enabling production at a larger operational scale with a focus on safety, consistency, and performance.
Editorial context：The deployment of Dexterity's Foresight world model and Mech trailer loading systems at the FedEx Hagerstown Hub marks a significant step in scaling physical AI for industrial logistics. This collaboration demonstrates the integration of real-time perception, planning, and execution in high-volume operations, with a focus on safety and efficiency.
NVIDIA introduces Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics. It distills the capabilities of Cosmos-H-Surgical-Simulator into a causal, few-step student model, enabling interactive environments for policy evaluation and synthetic data generation. The system runs on a single NVIDIA RTX PRO 6000 GPU and supports integration with surgical platforms like Versius.
Editorial context：NVIDIA's Cosmos-H-Dreams represents a significant advancement in real-time generative simulation for surgical robotics, enabling interactive training and evaluation of vision-language-action policies. It builds on prior work with Cosmos-H-Surgical-Simulator and introduces a causal, few-step student model optimized for real-time performance.
NVIDIA has released an open-source, GPU-accelerated Medical Physics Simulation framework as part of its Isaac for Healthcare platform. This tool enables developers to model anatomy-device interactions, generate complex scenarios, and train robot policies in simulation, significantly reducing development time and improving regulatory readiness.
Editorial context：NVIDIA's open-source Medical Physics Simulation framework integrates classical physics and generative AI to enable realistic medical robotics training, offering scalable simulation environments for anatomical and device interactions.
MIT's JARVIS Challenge explored whether AI can transform complex engineering tasks like building jet engines. Students used AI tools to design, fabricate, and test small gas turbine engines, but found that human expertise and judgment remained critical. The challenge revealed that while AI can speed up design and testing, physical manufacturing and engineering experience are still essential. The study emphasizes the need for a balance between AI assistance and human oversight in engineering workflows.
Editorial context：The JARVIS Challenge at MIT demonstrates how AI can assist in complex engineering tasks like building jet engines, but highlights the critical role of human judgment and expertise. The study shows that while AI can accelerate design and testing, physical manufacturing and engineering experience remain essential. The balance between leveraging AI tools and maintaining human oversight is key to safe
MIT researchers have developed SceneSmith, a system that uses AI agents and vision-language models to generate highly realistic and detailed virtual environments for robot training. These environments allow robots to practice tasks in simulated settings, reducing the need for physical testing. The system can create complex 3D scenes with a high density of objects, ensuring physical accuracy and realism, which helps improve robot performance in real-world scenarios.
Editorial context：SceneSmith represents a significant advancement in generating realistic, diverse, and physically accurate virtual environments for robotics training. By leveraging AI agents and vision-language models, it enables the creation of complex 3D scenes that closely mimic real-world settings, reducing the need for extensive physical testing and improving the efficiency of robot training.
Engineers at MIT and EPFL have developed a flapping-wing robot, or FAAV, that can swim underwater and then flap into the air, mimicking the movement of diving birds. The robot, weighing less than 300 grams, can transition between water and air using optimized wing size, flapping frequency, and tail angle. The study, published in Science, could enable new aerial-aquatic drones for oceanographic research and environmental monitoring.
Editorial context：This research introduces a flapping-wing robot inspired by diving birds, capable of transitioning between swimming and flying. The study highlights the biomechanical adaptations of diving birds and how they inform the design of a robot that can operate in both air and water without the need for feet. The findings could lead to new aerial-aquatic drones for oceanographic research.
NVIDIA and Hugging Face are releasing the NVIDIA Isaac GR00T 1.7 model and Isaac Teleop framework into LeRobot, an open-source robotics library, to provide developers with shared tools for training, evaluating, and deploying robot foundation models. NVIDIA Cosmos 3, a frontier world model for physical AI, is also set to be integrated soon.
Editorial context：NVIDIA and Hugging Face are collaborating to integrate advanced models and frameworks into LeRobot, an open-source robotics library, to streamline end-to-end robot development and foster community innovation.
Firefly Aerospace's Blue Ghost Mission 2 will deploy NVIDIA Jetson for on-orbit AI processing, enabling real-time lunar data analysis and reducing latency. This marks the first use of Jetson in lunar orbit, supporting scientific research and future lunar exploration.
Editorial context：NVIDIA Jetson's deployment in lunar orbit marks a significant step in edge AI for space applications, enabling real-time data processing and reducing reliance on Earth-based computation.
Dexterity's research introduces 'Transactable World Models' as a core component for Physical AI, treating world models as operators that reason about physical reality rather than storing data. These models enable robust manipulation by integrating physics, handling uncertainty, and supporting multi-agent coordination with explicit rollback and transaction guarantees. The approach emphasizes interpretability, real-time consistency, and the ability to reason about cause-and-effect relationships in dynamic environments.
Editorial context：Dexterity introduces 'Transactable World Models' as a foundational component for Physical AI, emphasizing interpretability, physics integration, and real-time consistency. This approach enables robust manipulation in complex environments by treating world models as operators rather than data stores, allowing for explicit uncertainty quantification and transaction guarantees essential for multi-ag,
Dexterity and ASRock Rack have partnered to develop custom AI supercomputing systems for Dexterity’s Mech industrial superhumanoid robot, enabling real-time edge intelligence for warehouse automation tasks. These edge AI servers are optimized for computational demands, supporting complex reasoning and manipulation at the edge without cloud dependency.
Editorial context：Dexterity and ASRock Rack's collaboration introduces custom edge AI servers tailored for industrial superhumanoid robots, emphasizing real-time processing and reduced reliance on cloud connectivity.
This hands-on tutorial walks through the process of collecting data, training policies, and deploying autonomous medical robotics workflows on real hardware using NVIDIA Isaac for Healthcare. It introduces the SO-ARM starter workflow, which enables developers to build and validate surgical assistant robots from simulation to deployment.
Editorial context：This tutorial provides a comprehensive guide to building a healthcare robot using NVIDIA Isaac for Healthcare, covering data collection, simulation, training, and deployment on real hardware. It emphasizes the use of simulation to generate synthetic data and the integration of real-world data for training policies that generalize across domains.
