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!
The IN²Bot team from the School of Automation at Beijing Institute of Technology won championships in embodied intelligence competitions at two major international conferences: it took first place in the CVPR 2026 ManiSkill-ViTac Visuotactile Embodied Dual-Arm Manipulation Challenge with a 700-point lead, using a solution that includes a self-developed tactile encoder and an action memory module; and it won the ICRA 2026 Fifth Mobile Robot Autonomous Navigation Challenge with the highest score overall, using methods covering a spatial traversability cognitive reasoning field, local reactive planning, and safety-barrier trajectory control.
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.
Boston Dynamics has released a new generation of dexterous hands for the Atlas humanoid robot, moving from 7 degrees of freedom in the previous generation to 13 degrees of freedom across four fingers, using a single actuator type for direct drive and full encapsulation. The new design is aimed at manipulation rather than simple grasping: a four-DOF opposable thumb supports fingertip sliding, pinching, three-finger grasping, and gripping tools such as drills and welding torches, while the strength still allows it to carry a mini-fridge payload of 100lb+, and it supports high-fidelity simulation for sim-to-real reinforcement learning.
A teleoperation demonstration of the hand of Torso 3.
Boston Dynamics has unveiled a dexterous hand for its next-generation Atlas humanoid robot, increasing from GR2's 7 degrees of freedom to 13, using a four-finger design with a 4-DOF opposable thumb, direct drive, and only a single actuator type. The design drops the pinky finger to reduce cost and complexity, is equipped with fingertip and palm pressure tactile sensors, supports pinch grasping, three-finger grasping, and pressing tool triggers, and was optimized for high-fidelity simulation for sim-to-real transfer and reinforcement learning, with preliminary sim-to-real results on dynamic tasks already available.
Why it matters: The original article details the degrees of freedom, actuation, and simulation design tradeoffs of the next-generation Atlas dexterous hand, letting readers understand how the hand design serves reinforcement learning and real-world work.
IROS 2026 was held in Pittsburgh, USA, from September 27 to October 1, with papers, workshops, and exhibition discussions focusing on how touch is entering robot learning, fine manipulation, and real-world applications. A paper from a University of Chinese Academy of Sciences team achieved a 67% success rate in a 0.05 mm clearance insertion task, reducing peak interaction force by 60% and torque by 44%; the experiments used PaXini PX6AX GEN3 sensors.
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.
STIAL Robotics recently won an AI flexible grinding project from a leading company in China's rail transit sound barrier industry. The grinding targets sound barrier columns up to 9 meters long. In the solution, the SeeV vision system uses 3D line laser scanning to extract weld seams, and NextBrain™ AI automatically matches grinding trajectories and force control parameters. The end effector is equipped with the PolishX flexible force control system, achieving contact force control accuracy of ±0.5% FS, capable of removing 2mm weld spatter residue and adapting to changeover between multiple column specifications.
STIER's AI flexible grinding robot has won another bid for a flexible grinding line equipment upgrade project at a well-known domestic medical device manufacturer. The workstation has run continuously for tens of thousands of hours on a real high-load production line, achieving an oxide layer removal rate of over 99%, raising pre-coating pretreatment yield from 85% to 98%, shortening per-piece processing time by 42%, and achieving contact force control accuracy of ±0.5% FS. Through its modular product system, AI process knowledge base, and digital twin offline verification, the company enables replicable, scalable delivery from single machines to complete lines.
In an interview, Yuansheng Intelligence founder Yang Sicheng proposed that when model capabilities are not strong enough, hardware capabilities must be very strong, and only after models break through can dexterous hands simplify. Its self-developed Apex Hand uses an in-palm hybrid tendon-driven scheme, has 21 degrees of freedom, a maximum single-hand active payload of 30 kg, and an open-close time of about 0.2 seconds, and was carried by the gold, silver, and bronze medal teams in the heavyweight categories of weightlifting and tug-of-war at the 2nd World Humanoid Robot Games in August 2026. He emphasized prioritizing reliability and dexterity, using unidirectional tendon drive plus end encoders to compensate for creep, and believes the endgame for dexterous hands is a reliable general-purpose productivity tool.
X Square Robot has released the TwinDEX system, consisting of a wearable embodiment-free data collection kit, a matching robot end effector, and data processing and policy training pipelines. Using hundreds of purely embodiment-free data samples (none from a real robot), the robot completed a chemistry experiment in a single continuous take, performing 24 sub-actions including opening a bottle and sampling, dropper operation, guiding liquid with a glass rod, and shaking to mix.
Mech-Mind will present 20 intelligent robot applications at the core of the robotics exhibition area at the 26th China International Industry Fair, held October 12–16 at the National Exhibition and Convention Center in Shanghai. Booth 7.1H-A225, with 300 sq m of exhibition space. On-site demonstrations will include dual humanoid robots working together on loading/unloading and bin handling, sorting of massive numbers of objects, and generalized grasping of transparent objects, covering more than ten industries including automotive, new energy, and 3C/semiconductors. Its general-purpose eye-brain-hand systems have been deployed more than 30,000 units globally, serving 100+ Fortune 500 customers.
At IROS 2026, under the theme "From Touch to Intelligence," Daimon Robotics publicly demonstrated its tactile intelligence infrastructure on real hardware for the first time, covering three layers: the DM-Tac series of vision-based tactile sensors (more than 12 tactile modalities and over 110,000 sensing units), the Data-Nexus data collection system, and the Daimon-TWM world model released in August.
We are releasing Dyna-2.1, the first physical agent to achieve reliable super long-horizon whole-body autonomy. It combines our new semi-humanoid hardware with Da...
At Session 13 of the Intelligent Robotics and Advanced Manufacturing Lecture Series, Professor Gu Guoying of Shanghai Jiao Tong University gave a report titled "Research Progress in Soft Robots," sharing results in three areas: high-performance artificial muscles, rigid-soft coupled dexterous robotic hands, and flexible tactile sensing and wearable systems.
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.
Sharpa launched three products at IROS 2026: the D01 manipulation robot with integrated tactile sensing, featuring an end-effector speed over 10.5 m/s, repeatability of 0.2 mm, and electronic-skin tactile sampling at 100 Hz; the W02 fully tactile compact dexterous hand, with 21 active degrees of freedom, about 30% smaller than the W01, and fingertip force sensing from 5 mN to 30 N; and the AE01 high-fidelity tactile exoskeleton data glove, which uses 22 encoders for teleoperation and data collection and supports 256-level continuous tactile feedback per fingertip.
Simplexity Robotics has launched and open-sourced the world-action model SimpleWAM; the 6B-parameter version reduces inference latency on NVIDIA ThorX (T5000) from 181.23ms to 42.31ms, enabling high-frequency closed-loop operation on-device at nearly 24Hz.
Why it matters: The original article presents the on-device optimization path from 181ms to 42ms and four architectural designs, which can be transferred to understand how world-action models can meet high-frequency closed-loop control requirements.
Sharpa unveiled three fully self-developed new products on-site at IROS in Pittsburgh: the tactile-sensing dexterous manipulation robot D01, the dexterous hand W02, and the exoskeleton tactile data glove AE01. D01 has a maximum end-effector speed exceeding 10.5 m/s, repeat positioning accuracy of 0.2 mm, and is equipped with a full-body electronic skin; W02 integrates 21 active degrees of freedom and is about 30% smaller than W01; AE01 includes 22 encoders and 256-level tactile feedback per fingertip, for teleoperation and data collection. The IROS exhibition runs until October 1 Beijing time, at booth 514.
SIASUN will exhibit at the Shanghai CIIF from October 12-16, at Booth A239, Hall 7.1H, National Exhibition and Convention Center, and will launch the "On the Road to Smart Manufacturing, Seeking Fellow Travelers" interactive event on site, where participants can receive small gifts. Exhibits include the wheeled humanoid robot Songyi Agile/Power, the One Hub Yishu multi-robot collaboration system, a 9-axis collaborative welding system, the "Xingweilai" intelligent cleaning robot QD750F, and more than 10 other workstations.
Yifei Technology exhibited its first humanoid robot, Hongjun, and the Lobster wafer robot at the 5th Global Digital Trade Expo in Hangzhou from September 23 to 27. Hongjun demonstrated electricity meter assembly on site, featuring a 7-DOF single arm, a 0.6-meter reach, a 5 kg peak payload, and repeat positioning accuracy of ≤±0.05 mm, and is equipped with the self-developed YiBrain multimodal industrial embodied large model; Lobster meets the ISO Class 1@0.1μm cleanliness standard and has entered SMIC production lines. The company listed on the Hong Kong Stock Exchange in May 2026, with overseas revenue up 318% year on year.
Qualcomm Technologies has reached an agreement to acquire PickNik Robotics, the developer of the MoveIt open-source manipulation framework, and will advance the integration of MoveIt with the Qualcomm Dragonwing platform while keeping MoveIt open source.
A team at Carnegie Mellon University's Robotics Institute has developed the LAMP (Long-Horizon Adaptive Manipulation Planning) system, which helps multiple robots collaborate to move objects in severely cluttered spaces by jointly planning object and robot motion rather than handling them separately.
Simplexity Robotics showcased two embodied intelligence demos at the 2026 Yunqi Conference: logistics parcel handling and parcel tape sealing and packing, and played a video of a CNC loading and unloading application already in a real production environment. The two logistics demos cover, respectively, autonomous recognition, grasping, and posture adjustment of randomly placed parcels, and task understanding, continuous operation, and dual-arm collaboration. Simplexity Robotics is training and iterating its self-developed LaST embodied foundation model based on infrastructure including Alibaba Cloud's Platform for AI (PAI) and Zhenwu AI chips.
ETH Zurich's Soft Robotics Lab taught an off-the-shelf humanoid robotic hand to walk autonomously, press arrow keys, and push blocks. With a battery, motion sensors, and a microcomputer added to the back of the hand, and after reinforcement learning training in simulation, the hand crawled on 14 indoor and outdoor surfaces, and recovered its standing posture on its own in 21 of 25 sideways-fall tests. It pressed the correct arrow key 29 out of 32 times. In 15 tests, the block was pushed to within 2 centimeters of the target. The paper Fingers as Legs has been published on arXiv.
Zerith Robotics appeared at the 2026 World Manufacturing Convention in Hefei, showcasing embodied intelligence data, the Zerith V series action models, robot bodies, and unmanned retail applications. Its "grab anything" demonstration achieves autonomous judgment and grasping for small, flexible, transparent, and reflective objects, and its robot shops have been deployed at 20+ locations, with plans to expand to 300–500 within the year. In the 2026 first-half rankings released by MIR DATABANK in the third quarter of this year, Zerith ranked 13th in humanoid robot production and 14th in shipments.
Zero Power Robotics appeared at the 2026 CIFTIS as a representative company expanding embodied AI overseas; its robot convenience store project demonstrated an unmanned retail loop in which, after a customer scans a code to place an order, the robot autonomously completes product recognition, grasping, and delivery. Recently, multiple Zerith H1 wheeled-arm humanoid robots have been deployed at a laboratory in Malaysia for local solution providers to conduct secondary development, data collection, and model training. The company is continuously iterating its Zerith V series embodied-native motion models and advancing the API-ization of embodied AI physical grasping.
See you at IROS 2026 🤖 T1 overseas debut — one platform, infinite scenarios 🧠 Lumo-2, the next-generation world-action model — live demonstration of high-precision manipulation AI that grows through practice 📅 September 27–October 1 📍 Pittsburgh, USA 📌 Booth #921 #IROS2026 #PhysicalAI #Astribot
Zhijian Dynamics will present its full-scenario embodied intelligent robot i7 Pro and full-scenario service robot iX at IROS 2026 (Booth 228). The i7 Pro will demonstrate carton sealing and packing operations on site, while the iX will showcase teleoperation and WBC whole-body control capabilities. During the conference, there will also be a poster presentation on sub-millimeter-precision CNC lathe operation research and a Tech Talk titled "From Lab to Factory: Analyzing the Challenges of Real CNC Operations for Existing Robot Learning Pipelines."
MagicLab introduces the visual guidance approach in the Magic-Brain framework: a VLM Agent outputs a bbox, and a segmentation model refines it into a three-color mask overlaid on the observation image, converting semantic guidance into visual guidance.
Why it matters: MagicLab details the design motivation and implementation of Magic-Brain's visual guidance layer, letting readers understand the impact on grasp success rate of converting language instructions into pixel guidance.
On September 15, a delegation from CNOOC Energy Technology & Services Equipment Technology Co., Ltd. visited Shenzhen Ruoyu Technology Co., Ltd. The two sides discussed intelligent solutions for high-risk offshore oil and gas field operations such as inspection and operation, valve switching, sampling, and pin removal, focusing on explosion-proof dexterous hands, offshore inspection and operation robots, and data collection for special scenarios.
The AI for Industry Challenge, organized by Intrinsic with Open Robotics, Foxconn, Google DeepMind, NVIDIA, and Universal Robots, has ended. 5,000 developers from 115 countries participated, with a prize pool of $180,000.
The East China Sea voyage organized by Tsinghua University in collaboration with the Qingdao Institute of Marine Geology has returned successfully. Using the independently developed deep-sea controllable visual sampler (DCVS), the team completed dive operations at 7 stations in areas including the Xianchi cold seep area and the Qiaoyue hydrothermal field, with a maximum dive depth of 1260 meters. DCVS operated stably under extreme conditions such as high temperature and high pressure, collecting multiple physical samples and acquiring large amounts of high-definition seabed imagery. The equipment was developed under the leadership of the Department of Automation at Tsinghua University, with participation from Shanghai Jiao Tong University and Xiamen University.
Standard Robots' industrial embodied intelligent robot DARWIN-Lift participated in the Huawei Cloud Stellar Innovators Competition - Robot Labor Contest during Huawei Connect 2026, and was awarded the High-Reliability Sorting and Handling Award. The model uses an industrial-grade omnidirectional mobile chassis, dual-arm compliant control, and VLA on-device compute inference to achieve a closed loop of mobility, perception, and manipulation, with a vertical torso lift stroke of 900 mm, dual-arm 10 kg-class payload, and a working range diameter greater than 1880 mm.
At the 23rd China-ASEAN Expo, Yuequan Bionic demonstrated its wheeled humanoid robot BOWEN W-Bot writing Chinese calligraphy with a brush and its Y-Hand M2 playing rock-paper-scissors with humans, among other demos; the results were reported on CCTV News. BOWEN W-Bot is equipped with the bionic dexterous hand X-Hand M1, which has 472 tactile sensing units and, combined with full-body hierarchical neural network motion balance control, enables brushstroke force perception. Y-Hand M2 has 38 degrees of freedom and interprets gestures in real time with millisecond-level response; related products also appeared in the "With Hui Tongxing" livestream, attracting more than 200,000 online viewers.
Ken Goldberg of UC Berkeley argues in an article for the Agentic Robotics (AR) approach, in which multi-agent systems write, test, and iterate structured robot control programs offline, as a third path between model engineering and data-driven methods that does not rely on large-scale robot demonstration data.
STIAL has launched an AI flexible grinding solution for grinding weld scars, weld seams, and burrs on welded enclosures, covering sheet metal scenarios such as energy storage enclosures, computing power cabinets, and electrical enclosures.
Under the guidance of teachers Zhang Wenqiang and Zhao Qing, a Fudan University student team of Wang Jianan, Li Jinghui, and Jiang Kaixun won a team first prize in the business plan special track at the national finals of the 21st China Graduate Electronics Design Contest (held in Wuzhen, Tongxiang, Zhejiang) for their project "Fangwei Dudian: the world's first industrial-grade embodied intelligent advanced power operation and maintenance equipment."
Figure AI released Helix 2.5, which completed three long-horizon full-body behaviors—tidying a living room, folding towels, and making a bed—zero-shot in 30 Bay Area homes where no data had ever been collected, without collecting data or fine-tuning in the evaluation environments.
Why it matters: The original article provides controlled comparison data on Index pretraining for zero-shot generalization and evaluation details, helping readers understand the actual level of humanoid robot pretraining transfer.