Today's Highlights
· Samsung establishes robotics division RX reporting directly to the CEO, poaches former Boston Dynamics strategy lead
· AGIBOT (Chinese humanoid robotics startup) reportedly pushing for IPO at a $20 billion valuation, with revenue projected at RMB 4 billion this year
· Nvidia open-sources Cosmos 3 Edge: a 4B-parameter world model that generates actions directly on-device on robots
· Momenta secures Robotaxi road-test permit in Shenzhen, stock up over 10% in a single day
· Alibaba releases 122B embodied foundation model RynnBrain 1.1, beating closed-source models across three spatial reasoning leaderboards
I. Research Progress
RynnBrain 1.1: 122B embodied foundation model beats closed-source models across three spatial reasoning leaderboards · vla
Alibaba DAMO Academy has pushed its open-source embodied foundation model to 122B-A10B scale, currently the largest open-weight tier publicly available, with HF trending at 30↑. Compared to version 1.0, the full lineup adds contact-point prediction, and the 2B/9B variants gain native 3D grounding, making outputs more tightly aligned with actual mechanical manipulation. The accompanying RynnBrain-VLA uses a unified cross-embodiment action space plus embodiment-specific masks, and has already been deployed on three robot platforms: Unitree G1, Astribot-S1, and Tianji Wuji. The 122B-A10B model surpasses all closed- and open-source models evaluated in the paper on VSI-Bench, MMSI, and RefSpatial-Bench; in real-robot experiments, policies initialized with it also outperform the Qwen series and mainstream general-purpose VLAs.
Kehan Li et al. (Alibaba DAMO Academy) · arXiv 2607.17977 source
Qwen-VLA: Manipulation, navigation, and human first-person data folded into a single policy · vla
The Qwen team is targeting a fragmentation problem: a dual-arm robot that can cook can't walk, and vice versa—manipulation arms predict joint poses while navigation robots predict waypoints, and the two have historically been hard to share in one model. Their approach uses Qwen3.5-4B as the cognitive backbone connected to a DiT flow-matching action decoder, then writes robot morphology and control mode into the input text via "embodiment-aware prompts," sharing the same latent space without changing the architecture. As a single generalist, it achieves 97.9% on LIBERO, a 76.9% out-of-distribution generalization success rate on a real dual-arm ALOHA platform, and 56.7% on RoboCasa-GR1, surpassing models specifically fine-tuned for that platform.
Qwen Team · arXiv 2605.30280 source · Analysis: Embodied Intelligence and Spatial Perception source (WeChat, CN)
Patch Policy: Dense visual features without the VLM baggage · manipulation
This work, with Yann LeCun as a contributor, points to an awkward tradeoff: robot policies either compress each frame into a single global token—losing spatial detail—or shoulder the full inference cost of a billion-parameter VLM. The authors simply add a block-causal attention mask, letting an ordinary transformer policy directly consume dense patch tokens from a pretrained ViT while preserving temporal causality. Across four simulation suites and three real-robot environments, this improves on global pooled representations by 40%; it outperforms fine-tuned OpenVLA-OFT by 18% while using only about 0.7% of its parameters.
Gaoyue Zhou et al. (NYU) · arXiv 2607.18236 source
POT-VLA: Giving humanoids memory that "the object is still there," success rate up from 39/80 to 71/80 · locomotion
A hidden failure mode in long-horizon humanoid loco-manipulation is "object state divergence": the object state used to drive whole-body actions and the state used to judge whether an action achieved its goal are actually two separate representations. The authors use persistent object tokens to maintain role-indexed 3D object records from RGB-D input, with the same record both conditioning action generation and supporting geometric predicate verification, closing the execution loop. Across eight real-world task categories on Unitree G1, the aligned GR00T-N1.7 baseline improved from 39/80 to 71/80.
Peng Ren et al. · arXiv 2607.18016 source
PolicyTrim: No retraining needed, just teaches the VLA to stop taking detours—up to 5.83x faster end-to-end · vla
What actually slows down VLA deployment is often not single-inference latency, but how many replanning cycles and real physical steps the policy needs. A Sichuan University team uses two-stage reinforcement learning post-training: first letting the model discover reliable action-chunk execution boundaries, then using a step-saving reward to suppress correction and retreat actions. On LIBERO, π0.5 achieves a 5.83x end-to-end speedup while maintaining over 98% success rate, with physical step count reduced by up to 51.4%; on a real Agilex Piper robot, the average speedup is 1.86x.
Sichuan University, Lei Yinjie's team · arXiv 2606.22540 source · Analysis: AI Era (Chinese AI media outlet) source (WeChat, CN)
Test-time scaling for world-action models: Zero-shot trajectory selection via geometric consistency · world-model
Robots can't compute an answer first and then choose, the way language models do—actions must execute immediately, so the model must judge on the fly whether spending extra compute is worthwhile. The authors exploit the fact that world-action models emit both action chunks and predicted future observations at each rollout, using a frozen geometric foundation model to compute cross-view depth reprojection consistency to rank candidates, requiring no training or task labels. RoboCasa's group average score rises from 66.3% to 68.4%; with consistency gating added, extra sampling is triggered at only 26.2% of decision points, recovering 74.8% of the gains achieved by the always-on mode.
Zesen Zhao et al. · arXiv 2607.17454 source
Adding a depth prior to JEPA world models lets an 18M-parameter model handle in-the-wild data · world-model
Complex, unpredictable outdoor real-world robot video has long been a challenge for JEPA-style world models. The authors introduce depth supervision plus isotropic latent-space regularization during training, over-parameterizing only at training time with no added inference cost. Training an 18M-parameter model on agricultural robot video, the frozen representation's error on a visual odometry probe is 33% lower than the LeWM baseline, and its "surprise" discrimination on out-of-domain TartanGround also clearly widens, with the advantage growing with rollout length.
Usman M. Khan · arXiv 2607.16314 source
"Reason before acting" isn't necessarily more stable—one class of VLA actually collapses fastest · vla
Intuitively, policies with reasoning steps should better absorb disturbances. The authors compare reasoning-free, text chain-of-thought, and latent iterative loop models head-to-head on LIBERO and SimplerEnv, and find the opposite: the latent iterative class collapses outright under random noise and white-box attacks, while the other two hold up, and adjusting reasoning depth barely helps recover it. Even more relevant for practitioners: a seemingly near-perfect "plan-action consistency" probe drops to random-chance level under adaptive attacks.
Tuan Duong Trinh et al. · arXiv 2607.17786 source
MEVION: A $14,000 open-source dual-arm data collection rig that fills ALOHA's gap in handling heavy loads · benchmark
ALOHA is already the de facto standard for dual-arm data collection, but its force and speed ceilings are low—heavy objects and fast motions are both off the table. The authors use sheet-metal welding to build a large-frame body from a small number of parts, all sourceable off-the-shelf; the four 6-DOF arms weigh 7.0 kg each with a peak torque of 60 N·m, and the full setup costs around $14,000. The elbow joint borrows a closed-chain mechanism from quadruped robots to reduce distal mass. Both hardware and software are fully open-sourced.
Kento Kawaharazuka et al. · arXiv 2607.17970 source
Other papers today: BoxTwin learns the full dynamics of elasto-plastic articulated objects from video, pushing digital twins toward deformable objects (arXiv 2607.17132 source); PhyAgentOS provides a unified runtime for heterogeneous embodied agents, separating "execution termination" from "semantic completion" (arXiv 2607.16636 source); World Translation uses inverse dynamics extraction plus unpaired domain translation to compress the sim2real gap (arXiv 2607.18154 source); FM-VLA adds force-sense memory to VLA for contact-rich manipulation (arXiv 2607.18231 source); Foresight Residual RL uses offline-estimated foresight value to optimize subtask handoff quality in long-horizon assembly (arXiv 2607.16506 source); an asynchronous multimodal diffusion policy handles the mismatch between each modality's perception rate and inference latency (arXiv 2607.17257 source); SinD 2.0 releases a multi-city signalized intersection drone dataset with annotated semantic risk (arXiv 2607.16943 source); A2RL Vmax provides a long-distance, high-speed perception and multi-vehicle interaction dataset for autonomous racing (arXiv 2607.17813 source); GeoWorldAD applies a geometric world-action model to autonomous driving planning (arXiv 2607.17521 source); UniETP unifies observation formats and action types across different simulators for general-purpose embodied task planning (arXiv 2607.18062 source); Token-Wise Latent Streaming lets a slow reasoner stream hidden states token-by-token to a fast planner, compressing observation staleness (arXiv 2607.16806 source); importance sampling plus PCA finds rare failures in a commercial autonomous trucking stack (arXiv 2607.18106 source).
Open Source · Tools · Benchmarks
· NVIDIA Agent Toolkit × Omniverse library: At SIGGRAPH, Nvidia folded its Omniverse library into the Agent Toolkit, letting AI agents build "simulation-ready" worlds directly, alongside 21 newly announced graphics and simulation research results source
· lingbot-world-2: Claimed to be the world's first hour-scale real-time world model, now open-sourced—moving the camera generates responsive scenes in real time source (WeChat, CN)
II. Funding and Deals
AGIBOT ｜ Planned IPO ｜ Target Valuation $20 billion (approx. RMB 150 billion) · humanoid ⚠️ unconfirmed report
The company has brought on CITIC Securities as underwriter, with full-year 2026 revenue projected at RMB 4 billion—already over RMB 1 billion in Q1 alone. The revenue trajectory underpins this valuation: RMB 300,000 in 2023, over RMB 60 million in 2024, and past RMB 1 billion in 2025. Unlike peers betting on research and demonstration shipments, AGIBOT has focused on industrial manufacturing—Longqi's Nanchang production line has deployed in parallel, with Furian Precision and SAIC factory lines running continuously; cumulative units shipped topped 10,000 as of March this year. Co-founder Peng Zhihui has previously said the company's biggest current shortage isn't capital but data.Source: Robo Encyclopedia (Chinese robotics industry outlet) source (WeChat, CN) · The Standard source
GigaAI ｜ New Round ｜ $3 billion valuation ｜ Planning Hong Kong listing this year · world-model
Following three rounds—Pre-B, B1, and B2—totaling RMB 3.5 billion between March and June, founder Huang Guan has publicly stated discussions are underway for a Hong Kong listing as early as 2026, aiming to be the "first world model IPO." The technical anchor is its proprietary embodied world model GigaWorld-1, which topped Tsinghua's WorldArena benchmark in April with a score of 62.34—the only embodied world model on that leaderboard to break 60 overall. Commercially, products are already deployed at FAW's die-making operations and Alibaba Cloud, with the general-purpose robot Maker H01 targeting delivery of a thousand units this year.Source: The Standard source · Robo Encyclopedia source (WeChat, CN)
Shouxing Technology ｜ Series A2 ｜ Hundreds of millions of RMB · embodied
Led jointly by CITIC Goldstone and Geely Capital, with CITIC Securities International Capital and Guotai Junan Innovation Investment following, existing shareholders Zhongwang Capital and AGIBOT increased their stakes, and Lei Jun's Shunwei Capital oversubscribed. The company specializes in bionic faces that can blink, frown, and smile, pursuing an emotional-interaction path rather than competing in whole-body humanoid platforms. Funds will go toward expanding its bionic product line and building standardized mass-production lines.Source: Tech Capital Circle (Chinese venture funding outlet) source (WeChat, CN)
Striverobotics ｜ Series C ｜ Hundreds of millions of RMB · industrial
Led jointly by Zhejiang state-owned Qianggang Capital and global optical module leader Innolight, with Guangtou Capital, Guohai Securities, and Juntong Capital following, and existing shareholder Fosun RZ Capital increasing its stake. The company builds embodied robots for high-altitude specialized work—a niche overshadowed by the humanoid boom but with among the most rigid demand for replacing dangerous human labor.Source: Tech Capital Circle source (WeChat, CN) · 36Kr source
Sereact (Germany) ｜ Series B extended to $116 million ｜ Zalando takes strategic stake · industrial
European e-commerce giant Zalando has joined this Berlin-based warehouse robotics company's Series B as a strategic investor, with Headline leading; total funding raised now exceeds $145 million. More than 200 Sereact systems are currently running across Europe, with customers including Mercedes-Benz Trucks and BOL. The funding will primarily go toward its next-generation robot brain, Cortex 2.0, whose approach is to use a learned world model to predict physical outcomes in advance and score candidate trajectories by risk and efficiency—shifting warehouse automation from "try it and see" to "compute first, then act."Source: Retail Technology Innovation Hub source
T2 (Japan) ｜ New Round ｜ Daiwa House Ventures and Mitsui Fudosan invest · autonomy
This Japanese L4 line-haul trucking logistics company secured investment from two real estate and logistics real estate giants after completing 500 km of real-world autonomous truck testing—the identity of the investors is itself a signal: Japan's commercialization entry point for driverless trucking is being staked on logistics real estate.Source: International Business Times sourcesource
Robopoet ｜ Strategic Investment ｜ Tuya Smart · adjacent
Following a Pre-A round of over RMB 100 million this month, IoT platform Tuya Smart has entered as a strategic investor, pointing toward mass production and distribution channels for consumer-grade embodied products.Source: PR Newswire source
Industry data: Global robotics startup funding hits $18.8 billion in H1, already surpassing all of 2025 · adjacent
A Crunchbase report shows global robotics startup funding reached $18.8 billion (approx. RMB 136 billion) in H1 2026, far exceeding all of 2025's $15 billion—a historic high. The trend is even steeper in China: by IT Juzi's count, embodied intelligence funding in China reached RMB 93.5 billion in H1, roughly 5x year-over-year, with 322 deals, up 137% year-over-year, and the number of unicorns valued above RMB 10 billion rising from 3 in 2025 to at least 26. In Hong Kong, 10 companies with robotics businesses have already listed in H1 this year with 16 more in the pipeline—yet the two that listed earliest, Huayan Robotics and Ledong Robotics, have both seen their stock prices fall by nearly half from their highs, suggesting secondary-market price recalibration has already begun.Source: Medical Device Business Review (Chinese medtech outlet) source (WeChat, CN) · Robo Encyclopedia source (WeChat, CN)
III. Commercialization and Deployment
Tesla expands Robotaxi to Tampa and Orlando; Cybercab to feature built-in Starlink antenna · autonomy
Following its expansion to Miami this month, Tesla has pushed Robotaxi service to two new Florida cities, Tampa and Orlando, ahead of its Q2 earnings report. Also announced the same day: Cybercab will integrate a Starlink V5 antenna—for a driverless fleet, remote takeover and telemetry links in cellular dead zones have long been an operational hard constraint, and satellite backhaul fills that gap. Robotaxi and Optimus progress are also expected to be the focus of Wednesday's earnings call, with markets watching whether cash burn can translate into a per-vehicle profitability model.Source: Reuters source · Electrek source
Momenta Robotaxi secures road-test permit in Shenzhen, stock up over 10% in a single day · autonomy
Shenzhen becomes Momenta Robotaxi's fourth core city in China after Shanghai, Suzhou, and Wuxi, with overseas testing already underway in Munich, Germany and Abu Dhabi, UAE. The market reaction was immediate, with the company's stock up over 10% that day. A previously reported mass-production custom vehicle project with SAIC's Xiang Dao is still on track for a 2027 reveal, with SAIC manufacturing, Momenta providing the L4 solution, and Xiang Dao operating—a division of labor that shifts Momenta from a Tier 2 solution seller toward a deep partnership with the automaker.Source: Eastmoney source · Momenta source (WeChat, CN)
BMW locates humanoid robot software center in Landshut, prioritizing data and simulation before production-line deployment · industrial
BMW has designated its Landshut, Germany components plant as the central development site for humanoid robot software, handling training data generation and preparation, virtual simulation, motion planning, and robot training—splitting responsibilities from existing production-line pilots in Leipzig and Spartanburg, which validate specific applications while Landshut builds the technical foundation. The technical approach is an open modular architecture combining fixed programmed workflows with AI models, where a VLA model translates visual information and task descriptions into concrete actions; data collection uses cameras, motion-capture suits, and data gloves. Project lead Christoph Jagoda emphasized that the evaluation covers not just technical feasibility but which processes humanoid robots can genuinely deliver economic returns for. Partners include Athenyx Robotics and Landshut University of Applied Sciences.Source: Automotive Manufacturing Solutions source
National-Local Joint Center and Huawei launch embodied intelligence training-ground demonstration site, Linglong robots already deployed at Yanfeng production line · industrial
The National-Local Co-Built Humanoid Robot Innovation Center and Huawei unveiled an "embodied intelligence training-ground demonstration site" during WAIC, targeting two bottlenecks: scarcity of high-quality data and insufficient model generalization. Huawei contributes compute, storage, networking, and its CloudRobo embodied simulation-training platform, while the innovation center contributes robot hardware, data, models, evaluation, and deployment. One concrete result: Linglong-series robots incorporating the joint research have already been deployed at Yanfeng, completing real-world data collection and autonomous work verification on the shop floor—validating the transfer path from "training-ground training" to "model iteration" to "industrial field deployment." Chief scientist Jiang Lei said the demonstration site has made breakthroughs in four areas: large-scale real-world data sets, Ascend-native VLA models, cloud-edge-device coordinated architecture, and an end-to-end toolchain.Source: Huawei China source (WeChat, CN)
Incheon Coupang warehouse fire burns for 61 hours, hundreds of lithium-battery robots on the fifth floor a hazard concern · industrial
A fire at a Coupang logistics center in Incheon, South Korea took 61 hours to fully extinguish, possibly the longest on record in the area, with flames burning through six floors; hundreds of lithium-battery-powered robots on the fifth floor were at one point flagged as an explosion risk. On-site investigation points to melted roller shutter doors and extremely high fire load. This highlights an under-discussed boundary condition as warehouse automation scales up: densely packed lithium-battery-powered robots are themselves a concentrated fire load, and existing fire-safety designs may not have been built around that premise.Source: Seoul Economic Daily source · Sohu source
RoboSense forms alliance with Abu Dhabi's Origen, pushing embodied intelligence into the Middle East · autonomy ⚠️ letter-of-intent stage
RoboSense (HKEX: 2498) signed an agreement with Abu Dhabi AI company Origen during WAIC, covering spatial intelligence, autonomous driving, industrial inspection, and smart cities, with Origen handling regional delivery, solution integration, and localization. Both sides currently describe the arrangement as "evaluating pilot deployment and localization strategy," with no specific orders or installation figures yet.Source: Zawya source
IV. Industry Developments
Samsung Electronics establishes robotics division RX reporting directly to the CEO, poaches former Boston Dynamics strategy lead · humanoid
Following last week's reports that Samsung plans to build a humanoid robot base in Gumi, Samsung Electronics has consolidated its scattered robotics efforts into a new organization: the Robotics eXperience (RX) Business Promotion Office, reporting directly to CEO Roh Tae-moon, overseeing mid-to-long-term strategy, core technology development, and commercialization, based at the Seoul R&D campus. Personnel moves are the strongest signal here—Executive Vice President Lee Dong-geun, who previously led robotics strategy at Hyundai Motor Group and Boston Dynamics, has been named head of the robotics strategy team, joined simultaneously by Seoul National University professor Kim Hyun-jin (intelligent autonomous robotic systems) and Ajou University professor Kim Eui-gyeom (manipulator and manipulation technology). Samsung has explicitly named data as the key competitive edge in robotics, and will build a Data Factory in Gumi to collect and analyze real industrial environment data, along with research centers in the US, China, and Japan. Samsung's stock rose over 5% following the announcement, and Gumi's mayor has publicly pledged full support for Samsung's 19 trillion won robotics and AI investment.Source: The Korea Times source · Chosunbiz source
Nvidia open-sources Cosmos 3 Edge: 4B-parameter world model generates actions directly on-device on robots · world-model
The Cosmos 3 family gains its smallest tier: Cosmos 3 Edge, a 4-billion-parameter model built for edge deployment—about one-sixteenth the size of Super (64B), announced at GTC Taipei on May 31, with Nano sitting at 16B. Its target scenario is specific: machines in factories, warehouses, and hospitals need to run data-center-level performance on memory-constrained bodies. The architecture is a Mixture-of-Transformers dual-tower design: an autoregressive tower handles visual and text tokens for understanding and reasoning, while a diffusion tower handles visual, audio, and action tokens for prediction, generation, and neural simulation; both towers retain their own normalization layers and MLPs while sharing multimodal attention, allowing the model to reason about a scene before generating output. The reasoner uses a 2B dense transformer, following Qwen3-VL's image and video messaging conventions. The key to cross-embodiment support lies in action representation: a vehicle's ego pose, a camera's motion, and a robot arm's end-effector pose plus gripper state are unified into compact geometric vectors describing translation, rotation, and manipulation state, connecting control directly back to the visual structure of the scene.Source: MarkTechPost source · GIGAZINE source
Model S autonomous-driving death case ruled $240 million in damages, three state attorneys general seek to overturn · autonomy
Tesla lost a lawsuit over a pedestrian death involving Model S's autonomous driving system, with damages ruled at approximately $240 million; three state attorneys general subsequently sought to have the ruling overturned. Beyond the damages figure itself, what's worth watching is whether this becomes a reference point for allocating liability in assisted driving—as NHTSA is in the process of rewriting driverless vehicle rules around "capability" assessments, how courts allocate responsibility between human and machine will have a longer reach than any single payout.Source: IT Home (Chinese tech news outlet) source
Beijing's industrial robot output up 75.5%, service robot output up 2.3x in H1 · industrial
According to Beijing Municipal Bureau of Economy and Information Technology data: in H1, "AI+" initiatives drove output growth of 92.1%, 75.5%, and 2.3x respectively in integrated circuits, industrial robots, and service robots. Over the same period, the city's combined industrial and information software sector value-add approached RMB 1 trillion, or 36.6% of GDP, with fixed-asset investment in key industrial sectors up 24.0% year-over-year, including a 36.0% increase in high-tech manufacturing investment. The gap between service robots' more-than-doubled growth rate and industrial robots' rate suggests this wave of volume growth is landing more in commercial service applications than on production lines themselves.Source: Beijing News source
Kexing Shikong, Shanghai Technology Exchange, and Unitree launch real-robot data demonstration platform · adjacent
A jointly built high-quality real-robot data demonstration platform has launched, with Kexing Shikong's data collection team having already deployed some Unitree robots to Fuxing Island in Yangpu District, collecting data on real-world tasks such as chair repositioning in retail settings. The notable element here is bringing a technology exchange into the embodied data pipeline—once real-robot data needs to circulate across institutions, pricing and rights allocation become unavoidable, an area that has largely been a blank slate until now.Source: Shanghai Technology Exchange source (WeChat, CN) · Kanchai.com (Chinese tech news outlet) source
Booster Robotics launches flagship humanoid Booster T2, powered by Nvidia Thor · humanoid ⚠️ manufacturer's own claims
Booster T2 debuted at WAIC, powered by Nvidia's Thor chip, with the manufacturer claiming 2,070 TFLOPS of compute; the company also launched a development toolkit, Booster Studio. The company's stated goal is moving humanoids from "can move" to "can be used"; no mass-production or delivery figures have been disclosed yet.Source: Daofengti (Chinese tech media outlet) source (WeChat, CN)
XPeng unveils TuringViT vision encoder · autonomy ⚠️ manufacturer's own claims
XPeng has unveiled a vision encoder for physical AI applications called TuringViT, claiming it restructures the training paradigm for large vision models. Only the launch announcement is public so far, with no third-party reproduction or benchmark comparison yet seen.Source: Gasgoo (Chinese automotive industry outlet) source
Hardware · Supply Chain
· Lingyi iTech · full-machine contract manufacturing: Beijing Yizhuang's embodied intelligence super-factory targets stable annual capacity of 10,000 humanoid units in 2026, rising to 20,000 in 2027; the Chengdu facility has already stabilized mass production of AGIBOT-customized units, while the Beijing plant handles Tiangong-series mass-production validation ⚠️ investor-communication claims source (WeChat, CN)
· Data gloves · a collection-side bottleneck: mainstream Manus gloves cost RMB 60,000–100,000 per set and prices are still rising while supply remains scarce; pairing with an RMB 100,000-class dexterous hand pushes per-set collection cost toward RMB 200,000–300,000, prompting BrainCo (a 21-DOF exoskeleton, full set RMB 160,000), Dexrobot, and Linjiedian to each build their own gloves source (WeChat, CN)
· Visuo-tactile sensing · lifespan is a real problem: once the silicone skin tears or glue leaks out, the entire fingertip's data is ruined and the part must be replaced and recalibrated; Linjiedian says it has reached a lifespan of 100,000 grasps, while some manufacturers report needing to recalibrate after just two to three days of intensive data collection source (WeChat, CN)
· Weihong Corporation · micro servo motors: claims its dexterous hand's micro servo motors overcome the stalling and burnout issues of coreless motors, with projected monthly output of 80,000 units by year-end ⚠️ manufacturer's own claims source (WeChat, CN)
· China Aviation Optical-Electrical Technology: has joined the Ministry of Industry and Information Technology's humanoid robotics and embodied intelligence standardization working group source (WeChat, CN)
Top comments (0)