Reading Modes
Open Artifacts with links to supporting source material.
782 selected reportsLast 30 days 706 recordsAll coverage 2,428 records
NVIDIA has released a fine-tuned GR00T-N1.7 model for apple pick-and-place tasks on a Unitree G1 humanoid robot. The model was trained on a 402-episode dataset with a 40,000-step cosine learning-rate schedule, and the checkpoint was saved at step 15,000. The model uses a flow-matching action transformer and is available in Safetensors format for commercial or non-commercial use.
VizMini2 is an Android app that allows users to view, drive, and inspect ROS 2 robots from their phone or tablet. It provides real-time visualization, teleoperation, and dashboard creation without coding, and works with Fast DDS. The app includes a demo for testing and is free with optional support through GitHub sponsors.
On October 12, 2026, the 'Industrial Embodied · Smart Manufacturing Without Borders' Industrial Embodied Intelligence Industry Innovation Summit Forum was held at the National Exhibition and Convention Center. The forum unveiled the Physical AI Industrial Embodied Operating System and the world's first Industrial Embodied Intelligence 6S Store, aiming to address technical bottlenecks and commercialization challenges in industrial scenarios for embodied intelligence.
Real2Gym is a research initiative proposed by institutions such as the Shanghai Artificial Intelligence Laboratory, aiming to construct interactive simulation environments through video, enabling robots to learn and transfer skills to the real world. This method combines geometric constraints, interaction keyframe correction, and physics engine validation, significantly improving task success rates and efficiency while reducing model invocation frequency.
MagicAtom's EchoPolicy-0 framework achieved a 75.4% average success rate in MolmoSpaces Combined benchmark, ranking first globally, while MagicBrain achieved 72.4% with a 94.4% grasp success rate. Both frameworks address challenges in VLA by integrating visual language models with geometric tools and action policies, improving task execution and generalization in complex environments.
Huiling Tech launched the eHand-20 dexterous hand at the 2026 China International Industrial Fair, priced at 9999 RMB. This 20-degree-of-freedom dexterous hand combines high degrees of freedom with low cost through a full direct-drive control solution, modular design, and scalable manufacturing capabilities. The eHand-20 not only enhances fine manipulation capabilities but also advances embodied intelligence through a hand-brain closed-loop system, marking a significant turning point in the transition of dexterous hands from laboratory research to industrial application.
Zhi Fangping, a Chinese embodied intelligence company, has achieved a valuation of 20 billion yuan and released its NeuroVLA model, a class-brain-inspired architecture that improves robotic response times and efficiency. The company has also made progress in commercialization, including partnerships and deployments in multiple industries. The article discusses the technical and market implications of these developments.
Euler Wanxiang and the Hong Kong University of Science and Technology (Guangzhou) jointly released UniWAM, a Unified World and Action Model, which enables unified learning of semantic understanding, future visual prediction, and action generation. UniWAM achieved state-of-the-art results on benchmarks such as LIBERO, LIBERO-Plus, and RoboTwin 2.0, and its capabilities in instruction following, complex interaction, and long-term task execution were validated through real-world experiments. The model code and weights are fully open-sourced, supporting further research and application.
Pi has released its latest robotic foundation models through the Pi API beta, enabling developers to fine-tune and deploy models for real-world applications. The API supports a range of industries and hardware, allowing companies to iterate on autonomy with reduced data and infrastructure requirements. Early deployments highlight the potential for practical robotic applications, with plans to continue updating models through the API.
This work introduces DexAgent, an agentic framework that transforms a single egocentric human video and task description into diverse robot training data. By reconstructing underlying physical interactions in simulation, verifying object properties, and generating optimized robot trajectories, DexAgent enables robots to learn complex manipulation tasks from one demonstration. Across eleven real-world tasks, policies trained on DexAgent-generated data achieve 63.6% average success compared with 18.2% for the strongest baseline. The framework organizes video-to-robot conversion into four stages: semantic understanding, object property-based simulation reconstruction, robot trajectory optimization, and robot data generation, supported by a shared tool library that grows across demonstrations.
This guide outlines the essential steps for qualifying SLS PA12-printed robot tooling for production environments. It covers defining the component's function, assessing load paths, designing for manufacturability, and implementing a staged qualification process. The tutorial emphasizes the importance of engineering validation, interface design, and testing under real-world conditions to ensure reliability and safety.
FineART is a bimanual manipulation dataset co-released by Scale AI and Hugging Face LeRobot. The data was collected and annotated by the Physical AI unit at Scale AI. FineART is a bimanual manipulation dataset where every episode comes with dense, human-written subtask annotations. It has 40,543 teleoperated episodes (1,718 hours), 533,913 subtask labels, and 151 tasks, all collected on a Trossen Stationary AI (ALOHA-style) bimanual robot.
UNITAS is the world's first robot task-native 3D world model, achieving cross-perspective and cross-body generalization capabilities by unifying observation, action, and environmental changes within a three-dimensional physical space. The model performs exceptionally well in multiple benchmark tests, including LIBERO, LIBERO-Plus, and RoboTwin 2.0. With just 1.7B parameters, it outperforms larger parameter models such as Fast-WAM (6B) and Motus (8B).
RoboFoundry is a new framework for embodied agents that treats the entire system as a self-evolving policy. By freezing the base model and allowing the surrounding system to evolve, it achieves significant improvements in task success rates across benchmarks like EmbodiedBench, RoboMemArena, and LIBERO-PRO. The system enables zero-shot transfer and online evolution on real robots, with results showing substantial performance gains compared to baselines.
NovaSplat is a model-free pipeline for 6D object pose estimation that trains a 3D Gaussian Splatting model from a short smartphone video. It serves as both a template for detection and a differentiable backbone for pose refinement, outperforming model-based methods on a household-object benchmark and model-free competitors on an industrial benchmark with textureless and specular surfaces. A faster inference variant, NovaSplat-Lite, is also introduced.
This paper introduces PartiGen, a part-aware latent diffusion framework that integrates anatomical structure into whole-body motion representations.
LIBERO-MAX is a study that evaluates the adaptability of robots in dynamic environments by designing paired tasks and mid-task changes to assess the performance of different strategies under environmental changes. The study found that the success rates of all tested strategies decreased, and the impact of different types of changes on the models varied significantly, indicating that robots need more robust perception and planning mechanisms to handle uncertainties in real-world scenarios.
Bai Ze Tong Jing is a less than six-month-old company based in Hefei, and its developed Causal World Model AWM enables robots to perform causal reasoning about environmental changes. AWM significantly improves data efficiency and task success rates by learning object, motion, and interaction patterns from unlabeled videos.
Agnes AI officially opens sources its agent execution infrastructure Agnes Harness (AGH) and device interaction protocol MHS (Model Hardware Standard), providing developers with a unified toolchain that supports the continuous execution of AI Agents in real-world tasks and physical devices.
NVIDIA has released a trained policy model for the Cosmos3-Nano robot, fine-tuned on LIBERO-10 data. The model achieves 94.4% success in standard scenes and 89.2% in perturbed scenes. It is part of a synthetic-data study and is loadable via the cosmos-framework policy server.
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).