The Embodied Data Atlas project has released v0.7.0-rc.1. The homepage adds animated scenes and the official DROID demonstration; six research-task cards show their top three entries and support Chinese and English search, quick filters, and shareable rankings. The catalog remains at 177 records and 171 independent families, includes 13,669 paper metadata records, passed 149 tests and 224 browser checks, and is deployed on GitHub Pages.
Researchers at Chemnitz University of Technology tested pedestrian reactions to an autonomous, manual, and teleoperated campus shuttle in three modes and at different approach speeds. Preliminary results show little difference in experience between modes, while faster approach speeds slightly reduce perceived safety and comfort. A separate experiment found that braking curves and stopping distance significantly affect how pedestrians perceive the interaction. The findings are preliminary and need to cover more vehicles and traffic scenarios.
The Embodied Data Atlas project has released v0.5.0-rc.1, adding six dedicated primary research task categories and expanding its reviewed catalog from 40 to 171 published independent families (177 records in total).
Researchers have developed a tiny magnetically controlled soft robot that can reach specific locations in the body under the guidance of an external magnetic field, leaving behind a small adhesive patch for precise drug delivery; the remaining robot can continue moving to the next target and repeat the placement. The technology could make targeted drug delivery more precise, but it is still at the research stage.
The author has released a daily-updated index of robotics papers, currently covering more than 480 papers from 2026, with an archive going back to 2017. Each entry includes links to the arXiv page, PDF, and code repository, and is tagged by topic for filtering. Paper pages provide plain-language summaries, key points, and figures from the original text, covering VLA, manipulation, locomotion, sim-to-real, humanoid robots, navigation, and reinforcement learning.
Researchers have developed a robot that can shoot artificial spider silk; the polymer solution forms hair-thin fibers in the air that can adhere to Teflon and bridge sharp broken glass. A single silk web supports 100 g, and a soft silicone gripper made from these fibers can pick up delicate flowers without damage.
Any2Track addresses the difficulty of a single reinforcement learning policy handling vastly different motions such as dancing, fighting, and getting up from a fall. It uses a two-stage approach: first training specialized teachers by cluster, then using DAgger to aggregate the teachers' behaviors into a single student network, and it presents G1 simulation comparison results.
Zhejiang University and other institutions propose Pumpire, a unified benchmark that directly evaluates the ability of image-level and video-level 3D foundation models to estimate point-to-point metric distances in reconstructed geometry, noting that previous approaches of separately evaluating depth and camera intrinsics cannot reflect this ability.
The GitHub project bhl-robustness-ladder has released the research-campaign-evidence-20261011 pre-release version, providing checksum-verified and validated source code and raw simulation evidence for preregistered follow-up research campaigns on perception, sensor fusion, and terrain. The project notes that published execution receipts do not imply a positive scientific result.
Khang Nguyen and co-author Minh Nhat Vu received the CASE 2026 Best Conference Paper Award for XFlowMP, a framework that uses Schrödinger bridges to transform random noise into trajectory distributions, generating feasible motions with velocity and acceleration based on start and end configurations.
ROMA proposes a real-world object-centric multi-sensor active perception system, in which an LLM planner determines missing information, actively collects it, and provides attribute estimates once evidence is sufficient. Unlike fixed pipelines that perceive first and then act, the system uses real multi-sensor hardware, and its code has been open-sourced.
At the tactile-themed workshop at IROS 2026, Columbia University's Yunzhu Li team presented progress roughly a year and a half after the open-sourcing of FlexiTac, an open-source flexible piezoresistive tactile sensor. The sensor uses a layered structure of FPC and force-sensitive film and can be assembled in 3 minutes after parts are prepared; the team open-sourced the complete BOM, assembly instructions, and reading software and integrated it into the LeRobot framework. Simulation adaptation requires calibrating only 3 hyperparameters of a spring-damper system.
Discrete Forcing proposes first using a single discrete forward pass to determine the overall structure of an action, then mapping it back to continuous space as the generation starting point for Flow Matching, completing fine correction through a single continuous forward pass, so that the entire action generation requires only 2 NFE.
Chelsea Finn points out that as robots become more reliable, DAgger-style data collection efficiency drops sharply; for example, a 90% success rate means a human spends more than 90% of the time as a passive observer. The Mulligan paper she cites proposes the key idea that failures often come from similar initial states, so learning can be focused on these states; in deployment, the robot performs the task and the human intervenes when needed, with each collection round focusing on the initial states where the robot fails. Across 3 real tasks and 2,550 blind tests, it outperformed uniform collection under the same budget. Paper: https://arxiv.org/abs/2610.05882 .
The Robotics and AI Institute (RAI) published a paper in Science Robotics introducing the dual-arm robot AthenaZero completing three baseball-inspired tasks: throwing a ball, catching a ball, and hitting a ball with a bat.
A team led by Professor Sejoon Lee at Dongguk University in South Korea combined a triboelectric nanogenerator (TENG) with a flexible graphene-channel ionic-gel-gated transistor to create a flexible device that needs no battery, harvests energy from human movement, and enables neuromorphic sensing and learning.
Leju, through the OpenLET community, has partnered with Qingtong Vision to release the LET-HumanMotion-Ruban2 dataset. Built on Qingtong's MotionDecode human motion capture data, it outputs standardized robot motion trajectories for Ruban 2 after retargeting, motion refinement, and format adaptation.
UCLA researchers have launched PhoneBot, a small open-source humanoid robot platform that uses an Android smartphone as its torso, handling visual perception, motion sensing, and motion control computing. Its hardware cost is about 400 US dollars (excluding the phone).
An 18-day joint Western Pacific hydrothermal expedition by Tsinghua University, the Qingdao Institute of Marine Geology, Shanghai Jiao Tong University and other institutions conducted seven dives and identified a new large high-temperature active hydrothermal field, preliminarily confirming its potential for large-scale polymetallic sulfide resources, with associated gold content of up to 15.4 ppm and associated silver content of up to 1,271 ppm.
Amirhossein Kazemipour's team at ETH Zurich's Soft Robotics Lab fitted a commercial robotic hand from Wuji Technology with an 80-gram backpack (containing a battery, IMU, and Raspberry Pi Zero) and used reinforcement learning in simulation to train a policy that lets it crawl on its fingertips.
Researchers at Oak Ridge National Laboratory's (ORNL) Manufacturing Demonstration Facility (MDF) are designing and validating smart automation systems that can sense their environment in real time and respond to changes, aimed at high-mix, low-volume production needs to expand the capacity of a constrained workforce.
In a white paper with IEEE Spectrum and Wiley, Noitom Robotics presents the HiPHI dataset, which contains 617.5 hours of optical motion capture full-body human motion data at sub-millimeter accuracy.
CMU researchers designed human-hand-inspired fingertips that can be mounted on the DeltaHands flexible robotic hand. They combine strain gauges to measure pinch pressure with a microphone that captures micro-vibrations through a plastic fingernail, enabling the hand to count paper cups, identify materials, and grasp fragile items such as tofu cubes and potato chips. Each fingertip costs less than 100 US dollars, compared with about 800 US dollars to build a similar setup from off-the-shelf products. The work was presented at the 2026 Intelligent Robots and Systems Conference in Pittsburgh.
Why it matters: The original text provides details on the multimodal tactile design of the human-inspired fingertips and their cost of less than 100 US dollars, helping readers understand how force and vibration sensing support manipulation control.
A University of Glasgow team used AR to overlay and test 5 prototype faces (animal, robot, emoji, anime, and eyes-only) and 5 sounds (cat meow, human-like voice, electronic tone, music, and "animalese") on the social robot Qoobo, with 24 volunteers ranking the combinations. Results showed that animal and anime faces paired with animal sounds were most liked, and that faces conveyed emotion better than sound. The study is based on their AZRA system and could inform the design of future zoomorphic social robots and user-customizable appearances.
SimForcing proposes a simulation-guided framework that injects motion knowledge from a simulation teacher into an action-conditioned robot world model through latent-variable distillation, mitigating appearance differences and reducing the extent to which inaccurate simulation predictions mislead real-video generation.
A research team from UC3M and UPNA published a paper in Bioinspiration & Biomimetics proposing a modular soft gripper that can roll and reorient objects between its fingers without releasing them.
Almost all dual-arm robots are equipped with wrist cameras, but these add bulk, cost, and complexity. EyeRobot 2.0 instead uses an active gaze approach with a head-mounted stereo camera to replace wrist cameras. —Do dual-arm robots really need wrist cameras? Maybe not!
Sanctuary AI proposes TIGER (Task-Space Imitation Guidance for Efficient Reinforcement Learning), which does not execute the imitation policy but instead maps its action chunks through the controller into end-effector reference trajectories, converting them into dense rewards for reinforcement learning.
Why it matters: The original text provides the specific approach for converting imitation policies into dense rewards along with ablation data, allowing readers to judge whether it can be applied to their own sparse-reward manipulation training.
Chemnitz University of Technology and TU Berlin tested pedestrian interaction with an autonomous shuttle from July to August 2026. Initial results show the interaction was generally perceived as safe and comfortable, with no significant effect from the three simulated operating modes: autonomous, manual, and teleoperated. Higher vehicle speeds slightly reduced perceived safety and comfort while increasing perceived efficiency; braking method and final stopping position also significantly affected pedestrian evaluations. Data is still being analyzed and will be published, with results to be presented on October 6 at CVAG
The paper LATENT, with Galbot as a participant, received the IROS 2026 Best Entertainment and Amusement Paper Award. The work starts from imperfect human motion data and achieves real humanoid robot tennis rallying. The post provides the project page http://zzk273.github.io/LATENT/, the arXiv address http://arxiv.org/abs/2603.12686, and the code repository http://github.com/GalaxyGeneralRobotics/LATENT, and congratulates the co-authors and collaborators.
Researchers at Georgia Tech have proposed an evaluation benchmark called FindingDory, using 60 tasks that require sustained interaction and awareness of environmental context to test the memory capabilities of vision-language models such as Gemini 2.0-flash and GPT-4o; the tasks cover three categories of high-difficulty scenarios: spatial, temporal, and multi-objective.
The ML3DHS 3D hierarchical semantic segmentation framework, developed by Assistant Professor Zhao Na's team at the Singapore University of Technology and Design (SUTD) with collaborators from Southwest Jiaotong University, was presented at the 43rd International Conference on Machine Learning (ICML) in 2026, and assigns each 3D point multiple related labels at once, from "furniture" to "chair" and "table".
Aaron Ames, director of Caltech CAST, believes safety is the most critical yet missing element before humanoid robots enter society, and that the lack of clear safety standards will inevitably lead to injuries; he notes that current demonstrations are mostly performance-driven, and few companies are seriously considering safe deployment.
At the IROS 2026 Tech Talk, Zhijian Dynamics presented its precision manipulation work for robot loading and unloading of CNC lathes, where the clearance between the workpiece and the chuck is only 0.5 millimeters; the approach covers motion generation, multi-view perception, historical memory, and contact feedback.
KAIST's quadruped robot RAIBO2 finished the Persimmon Marathon in Sangju, South Korea, in 4 hours 19 minutes 52 seconds, slightly faster than the human average, without swapping or recharging its battery, and still had 34% charge at the finish; the team described the work in a Nature paper last week.
Why it matters: The research moves energy saving for legged robots from optimizing single components to holistic design across joints and circuits, offering concrete methods transferable to long-duration outdoor robots.
Being-H0.7, the general-purpose dexterous manipulation large model independently developed by ZhiZaiWuJie, has been accepted to NeurIPS 2026, marking the Being-H series' completion of acceptances to the three top conferences ICML, CoRL, and NeurIPS within one year.
Professor Hong Chul Moon's team at KAIST published in Advanced Materials on Ionograsper, an ionic soft material inspired by the Venus flytrap that integrates sensing, actuation, and shape retention into a single material.
Researchers at the Chair of Intelligent Biorobotic Systems at the Technical University of Munich have developed a soft, lightweight exoskeleton that uses artificial tendons to assist hip extension, helping older adults stand up and walk more efficiently; the paper was published in Nature Communications.
CMU and Siemens partnered to develop a method for evaluating the "safe stoppability" of humanoid robots, and used simulation to define the conditions under which a robot can be safely stopped. The result is PRISM (Proactive Refinement of Importance-sampled Stoppability Monitor).
Why it matters: The article introduces PRISM, a method that uses simulation to define the conditions under which a humanoid robot can be safely stopped, and it is a useful reference for readers interested in humanoid robot safety monitoring and certification.
An EPFL and NYU team published ScaFi in npj Robotics, a biomimetic robotic fish based on the tail-swinging motion of cod and mackerel, scaled up by simply thickening a fiberglass tail rod proportionally. The team built three sizes—0.6, 1.1, and 2.9 meters—and tested them in a creek and Lake Geneva; after normalization by body length, the swimming gait was largely consistent. The largest model has lower energy efficiency and needs a stronger motor, while under disturbance the smaller models are more agile but recover more slowly.