category · newest first
Research
Japan's Human Augmentation Blueprint: Beyond Robots
Japan's Human Augmentation Consortium released a white paper proposing a unified framework for extending human physical and cognitive abilities. The document integrates technologies from major Japanese firms, aiming to create a standardized ecosystem for remote work and skill transfer.
Gartner Lists 10 Misconceptions About Physical AI
Gartner Japan has published a report identifying ten common misunderstandings about physical AI. The firm argues that companies must redesign their entire operational systems to remain competitive by 2030, rather than just adding robots to existing infrastructure.
InterMimicGen: Self-Evolving Motion Imitation for Humanoids
Researchers introduce InterMimicGen, a self-evolving framework that converts sparse human-object interaction data into executable humanoid robot motions. The system uses a data flywheel to iteratively refine and expand motion references, enabling a single generalist policy to handle diverse tasks.
Recursive Video In-Context Learning for Agentic Robots
The authors introduce RV-ICL, a method that structures demonstration videos into a multi-level hierarchy of sub-events. This allows LLM agents to access specific visual details during execution without slowing down the entire process. The approach improves success rates on LIBERO-PRO and LIBERO-Plus benchmarks.
Glasgow study tests AR faces for social robots
A University of Glasgow team used augmented reality to test various face and voice combinations on a Qoobo robot. They found that animal-like and anime-style designs were highly effective at creating emotional connections with users.
China Launches National Lab for Embodied AI Evaluation
China has activated a national-level system for evaluating intelligent robot performance. The Shanghai Robot Industry Technology Research Institute and the National Robot Testing and Certification Center have established a dedicated laboratory for embodied intelligence assessment. The facility has begun recruiting companies for a 27-item capability profile test that focuses on cognitive and interactive functions rather than just mechanical precision.
MIT's Light-Steered Muscle Robot Swims Through Water
MIT researchers demonstrated a chewing-gum-sized robot powered by a thin layer of living muscle cells. The device swims and steers in response to external light, achieving speeds of four body lengths per minute in a controlled maze.
Eli Lilly and Purdue to Present HRI Field Data at RoboBusiness
According to the report, Faraz Yousuf of Eli Lilly and Dr. Anastasia Kouvaras Ostrowski of Purdue University will present a session on human-robot interaction at RoboBusiness 2026. The talk will cover field learnings from studies involving automated guided vehicles in manufacturing and warehouse settings, as well as early findings on worker perception of robots.
First reported Oct 3
SNAP: Geometric Learning via Novel View Synthesis
The authors present SNAP, a self-supervised model that uses Novel View Synthesis to learn robust 3D geometric representations. By restricting the decoder's expressivity and using latent-space targets, SNAP outperforms standard methods across five robotics and vision tasks.
First reported Oct 3
EyeRobot 2.0 Uses Active Gaze for Bimanual Manipulation
The authors introduce EyeRobot 2.0, a system that performs fine-grained bimanual manipulation using a single stereo camera. By actively swiveling its viewpoint to center on 3D fixation points and allocating more computational resources to the image center, the system compensates for the lack of wrist cameras. In real-world trials, this approach outperformed passive stereo baselines by 40% and matched or exceeded ego-plus-wrist camera policies, particularly when objects occluded the wrist views.
Samsung unveils robot commercialization roadmap
Samsung Electronics has announced a strategic roadmap to commercialize robots, starting with factory automation and moving to home helpers. The plan relies on a new three-layer foundation model and a novel 'Metro-WM' world model designed to improve reliability and reduce costs.
Korea Forum Links Field Expertise to AI and Robotics
A recent industry forum in Gyeonggi Province highlighted the critical role of field expertise in AI and robotics. Speakers from Upstage, MaumAI, and KAIST argued that human judgment is essential for defining AI tasks and capturing skilled labor data for robot training.
SK Pack Leads 10B KRW Project for Physical AI Manufacturing
SK Pack has been selected as the lead agency for a Korean government-funded research project focused on physical AI-based autonomous manufacturing. The five-year initiative, backed by 10 billion KRW, aims to develop multi-agent AI systems that coordinate collaborative robots and autonomous mobile robots (AMRs) in real production environments.
Neuromeka and KAIST Develop Wrist-Camera-Free Manipulation AI
Neuromeka and KAIST have introduced SCDP, a physical AI technique that allows humanoid robots to perform precise manipulation without wrist cameras. The system uses a single head-mounted RGB camera to extract relevant visual features for bimanual tasks, outperforming traditional diffusion policies in simulation and real-world tests.
Jeonbuk University Selected as Core Hub for South Korea's 820 Billion Won Physical AI Project
Jeonbuk National University has been selected as a core research and validation institution for a major national Physical AI project worth 8.2 trillion won. The university will receive 265.18 billion won over five years to develop collaborative AI software, build a 190 billion won testbed, and establish an open research cluster.
First reported Oct 1
PrefPI Steers Robot Policies Beyond Initial Capabilities
Researchers introduce PrefPI, an iterative method that steers pretrained robot policies using only relative preferences over self-generated trajectories. The approach aims to push behaviors beyond the initial policy's effective support, where desired actions are rarely observed. In real-world tests, the method increased object transport height from 10.7 cm to 19.8 cm using just 150 preference-labeled trajectories.
First reported Sep 30
ASENA: Coding Agents Drive Robot Navigation
Researchers introduced ASENA, a system where general-purpose coding agents program robot behavior without retraining model weights. The approach integrates a 4B-parameter navigation policy, achieving top-tier results on standard benchmarks and improving performance through iterative self-correction.
First reported Sep 30
DSDyn-VLA: Dual-Stream Framework for Dynamic Manipulation
Researchers propose DSDyn-VLA, a dual-stream system that combines motion-aware planning with real-time correction to improve robot manipulation of moving objects. The authors report significant performance gains over existing methods in both simulation and real-world dynamic settings.
First reported Sep 30
LocoWM: World-Model-Guided Residual Adaptation for Locomotion
Researchers introduce LocoWM, a control framework that combines a base locomotion policy with a world model to predict future physical states. A residual adapter then generates corrective actions based on these predictions to improve precision and robustness.
First reported Sep 30
Adversarial Spheres Disrupt VLA Robot Models
Researchers propose a 'Universal Adversarial Object'—a sphere with optimized surface texture—that degrades the performance of Vision-Language-Action (VLA) models. The authors report that this object reduces average task success rates by 31.2% to 39.9% for two representative models, Pi0 and RDT, in both simulated and real-world settings.
MedInTech Showcases AI Endoscopy at ENDO 2026
MedInTech exhibited its INTION S robotic endoscope and INTION DEEPEYE software at the 5th World Congress of GI Endoscopy in Cape Town. The company reported that 89.3% of survey respondents rated the product experience as 'Excellent' and identified potential distribution partners in three African countries.
First reported Sep 30
VLA Failure Modes Under Camera Blackouts and Freezes
Researchers analyzed how vision-language-action models behave under camera faults like blackouts and freezing. They found that while task success rates are similarly low, the physical failure modes differ significantly, with freezing causing extreme joint behavior and blackouts leading to object drops.
KAIST Builds a Soft Robot Hand That Senses and Grabs with One Material
KAIST researchers developed 'Ionograsper,' a soft material that combines sensing and actuation. It detects charged objects without contact and grips them using UV light, maintaining the hold for over 10 minutes. The work was published in Advanced Materials.
Samsung AI Forum 2026: Agentic and Physical AI Updates
Samsung Electronics held its 10th annual AI Forum in Seoul, focusing on 'Agentic Shift: From Intelligence to Impact.' The event featured presentations on on-device AI, physical AI for robotics, and AI applications in semiconductor design and manufacturing.
NASA Accelerates Dexterous Robotics for Lunar and Mars Missions
NASA has introduced its Dexterous Robotics Team, which is developing hardware and software for robots to operate in extreme environments. The team is building iMETRO, a test facility that allows robots to interact with simulated spacecraft and habitats. The goal is to extend human capabilities in space rather than replace astronauts.
Daimon Robotics Unveils Tactile AI Stack at IROS 2026
Daimon Robotics presented a tactile-based physical AI technology at IROS 2026 in Pittsburgh. The company describes a three-layer stack that integrates tactile sensors, physical interaction data, and world models to allow robots to respond to forces, friction, and material changes in real time.
Samsung AI Forum 2026: Agentic AI and Self-Evolving Models
Samsung Electronics held its 10th annual AI Forum in Seoul, featuring keynotes from OpenAI and AWS on agentic AI. The company also presented research on physical AI, including world models and a 'self-evolving AI' system that improves based on user feedback.
Medin Tech Showcases AI Endoscopy at ENDO 2026
Medin Tech exhibited its AI robotic endoscope 'Intion S' and AI software 'Intion Deepeye' at the ENDO 2026 conference in South Africa. The company reported that over 200 attendees tested the devices, with a strong presence from the African region. Medin Tech plans to pursue distribution partnerships in several African countries based on the event's feedback.
Rho: Open-Weights VLA Models for Bimanual Robots
The Rho Team has released Rho, a family of open-weights Vision-Language-Action (VLA) models designed for bimanual manipulation. The authors report that Rho variants for YAM Box, UR AI Trainer, and FR3 Duo match or outperform existing open-weights baselines in both simulation and physical robot experiments. The models also demonstrate an online adaptation capability, requiring as few as 15 corrected episodes to handle tasks at the fringe of their training distribution.
First reported Sep 30
ReWAM: DINO-based world-action model for robot control
Researchers introduce ReWAM, a world-action model that leverages pre-trained DINO features to enhance robot policy learning. The model achieves a 93.6% success rate on RoboTwin 2.0 and an 8.28% success rate on RoboDojo, demonstrating strong performance without relying on generative video pre-training.
First reported Sep 30
EVO-WAM: Self-Improving Robot Policies via Video-Action Verification
The authors introduce EVO-WAM, a method that improves robot policies on new tasks without collecting additional expert demonstrations. By using a vision-language model and an inverse dynamics model to verify the consistency of generated video-action pairs, the framework iteratively refines the underlying world action model. Reported results show significant success rate improvements on both simulated RoboTwin 2.0 tasks and real-world composite tasks.
First reported Sep 30
EgoAlign: Adapting Human Demos for Humanoid Robots
Researchers introduced EgoAlign, a data-construction framework that transforms egocentric human demonstrations into supervision signals for humanoid robots. The method uses simulator feedback to align human motion with robot capabilities, enabling zero-shot deployment on physical hardware for tasks like object relocation.
IFR: Service Robots Enter Mainstream Industrialization Phase
The International Federation of Robotics (IFR) released a position paper stating that professional service robots are moving beyond niche markets into routine operational environments. The report identifies labor shortages as the top adoption driver and outlines five key sectors for deployment.
DexRoam: Learning Mobile Bimanual Manipulation from Human Demos
Researchers present DexRoam, a system that learns mobile bimanual dexterous manipulation from egocentric human demonstrations. By using a tracker-free capture setup and three alignment stages, the method preserves fine-grained whole-body motion, improving average success rates from 29% to 56% on GR00T N1.7 and from 32% to 57% on pi0.5.
Holo-M: Discrete VLA for Humanoid Loco-Manipulation
Researchers introduce Holo-M, a discrete Vision-Language-Action (VLA) model designed for humanoid loco-manipulation. It uses a unified action tokenizer and grouped discrete diffusion decoding to handle the high-dimensional action space of humanoids, achieving top success rates on the SIMPLE benchmark.
ScaFi: A Scalable Robotic Fish for Diverse Aquatic Environments
A team from EPFL and NYU has developed ScaFi, a robotic fish designed to operate in various aquatic environments without requiring a complete redesign for each size. The robot uses a flexible tail mechanism that scales proportionally, allowing it to maintain fish-like swimming motions from small creeks to large lakes.
UW Debuts 1g 'DirectHop' Robot with Precise Jump Control
University of Washington researchers have developed DirectHop, a 1-gram jumping robot that can control jump height in 1cm increments and self-right after landing. The design replaces traditional spring mechanisms with a direct-drive motor system to improve energy efficiency and precision.
NC State's Light-Powered Soft Robot Jumps Repeatedly
Researchers at North Carolina State University have developed a soft, non-electronic robot that jumps continuously when exposed to infrared light. The device uses a liquid crystal elastomer cord and a V-shaped aluminum stopper to store and release torsional energy, allowing it to reset itself between leaps. By adjusting the angle of the stopper, the team can control whether the robot crawls, jumps forward, or leaps vertically.
First reported Sep 26
JAXA Opens 7th Kibo Robot Programming Challenge Using Real ISS Robot
JAXA and NASA are running the 7th Kibo Robot Programming Challenge, letting student teams write autonomous flight software for the real Int-Ball 2 robot aboard the ISS Kibo module for the first time. Teams must guide the free-flying robot to investigate a simulated cabin pressure drop, with orbital finals set for spring 2026.
First reported Sep 26
General Robotics CTO Explains Modular AI Approach
The Robot Report podcast features an interview with Sai Vemprala, CTO of General Robotics, where he outlines the company's strategy for modular intelligence. The episode also covers recent industry news, including Agility Robotics exploring wheeled platforms and Qualcomm's acquisition of PickNik Robotics.