Today on The Robot Beat: GlobalFoundries is clearing floor space for the electric Atlas humanoid, marking Boston Dynamics' first major non-automotive pilot. Meanwhile, UC San Diego engineers hit record speeds with a 3D-printed soft robot powered entirely by pressurized air, and new vision-language-action models are bypassing pixel rendering to drive control latency down to 110 milliseconds.
Yesterday we covered Boston Dynamics and GlobalFoundries' expanded cleanroom deployment; today, supply chain disclosures reveal Boston Dynamics is preparing for an initial 10,000-unit Atlas production run, targeting a commercial unit price below $320,000. The early 2027 trials will place the electric Atlas in the same semiconductor sub-fabs where the company's Spot quadrupeds have already completed 90,000 inspections.
Why it matters
While the cleanroom deployment provides crucial non-automotive validation, the newly disclosed 10,000-unit production run indicates Boston Dynamics is moving beyond bespoke pilot manufacturing. A sub-$320,000 price point for a heavy-duty biped establishes a concrete capitalization benchmark for industrial operators replacing traditional factory automation.
While Boston Dynamics emphasizes cleanroom versatility, manufacturing analysts note that hitting a sub-$320,000 price point at a 10,000-unit scale will rely heavily on leveraging Hyundai's automotive supply chain efficiencies.
On Wednesday, November 18, researchers from Zhejiang University and Unitree Robotics open-sourced Movens-5M, a multi-modal pre-training dataset containing 5 million robot action trajectories spanning over 10,000 hours of behavior. Alongside the dataset, the team released Movens-Zero, a unified whole-body embodied AI model deployed on the Unitree G1 humanoid. The architecture incorporates motion expert modules scaling up to 910.5 million parameters, demonstrating that increasing trajectory volume directly improves multi-modal instruction tracking across text, speech, and video inputs.
Why it matters
Whole-body humanoid control has frequently been limited by fragmented datasets that separate upper-body manipulation from lower-body balance and gait generation. By providing a single open dataset and architecture validated across 10,000 hours of physical and simulated motion, Movens-5M establishes a public benchmark for whole-body policy pre-training. This enables independent developers to verify scaling laws on commercial bipedal hardware without generating proprietary data pipelines from scratch.
The Zhejiang University and Unitree research team asserts that open-sourcing multi-modal motion experts accelerates the development of generalist physical policies. Independent roboticists highlight that while Movens-Zero simplifies whole-body tracking, adapting the pre-trained weights to humanoids with distinct leg kinematics or mass distribution still requires fine-tuning.
On Thursday, November 19, Tesla CEO Elon Musk announced plans to release unedited, fully autonomous interaction videos of the Optimus humanoid robot within the next two months. The update comes as third-generation Optimus units—featuring 37 degree-of-freedom joints and redesigned dexterous hands—entered early mass production at the Fremont factory, reaching a weekly output of several hundred units. Production will eventually transition to a dedicated 7-million-square-foot facility currently under construction at Gigafactory Texas, aimed at large-scale manufacturing by mid-2027.
Why it matters
Persistent industry skepticism regarding teleoperation in public humanoid demonstrations has made verified autonomous execution a critical requirement for commercial credibility. Demonstrating unscripted, non-teleoperated tasks in unstructured environments provides a key benchmark as Tesla prepares for external home and factory leases. Concurrently, scaling early production to hundreds of units per week tests whether Tesla's proprietary joint design and supply chain can handle volume manufacturing.
Elon Musk maintains that Optimus will achieve basic household chore autonomy within 12 to 18 months, positioning physical AI as Tesla's primary long-term value driver. Financial analysts and robotics engineers remain cautious, pointing out that historical timelines for Tesla's full autonomy releases have experienced significant delays when moving from controlled pilots to real-world edge cases.
On Wednesday, November 18, Playground open-sourced Playground Serve, an internal inference runtime engineered to execute vision-language-action (VLA) robot policies on cloud-hosted GPUs. The platform supports 23 policies across 8 model families—including π0.5, MolmoAct2, and GR00T N1.7—utilizing CUDA graph recording and compiler-level operator merging. In benchmark testing, Playground Serve reduced compute costs by 64% to 72% compared to standard hosted endpoints while cutting first-token and chunk execution latency.
Why it matters
High inference latency and heavy cloud compute overhead represent major barriers to deploying large VLA models on physical robot hardware. By open-sourcing an optimized inference engine that runs directly within a developer's cloud or edge GPU infrastructure, Playground lowers the operational cost of running multi-billion-parameter policies. The capability to host models like GR00T and π0.5 on standard consumer hardware such as the RTX 4090 democratizes physical AI policy experimentation.
Playground Research asserts that self-hosted, compiler-optimized inference is necessary to eliminate vendor lock-in and high per-token API costs for physical robotics. Third-party developers note that while the runtime slashes execution costs, managing real-time network streaming between edge robot controllers and cloud GPU servers still introduces latency challenges over unreliable wireless connections.
On Thursday, November 19, developer apojomovsky released ros2serial v0.5.0, an open-source communication library that connects resource-constrained microcontrollers to ROS 2 nodes over USB serial. Written in pure C99 with zero dynamic heap allocation, the firmware consumes 11.8 KB of flash memory and 1.2 KB of RAM on an Arduino Uno. The v0.5.0 release implements a custom reversed COBS framing protocol and type-safe handshake mechanisms that validate REP-2011 type hashes across Arduino, ESP32, STM32, RP2040, and CH32V003 hardware.
Why it matters
Interfacing low-cost 8-bit and 32-bit embedded microcontrollers with high-level ROS 2 architectures usually requires heavy middleware or custom transport layers that exhaust microchip memory. By eliminating dynamic memory allocation and providing deterministic serial framing, ros2serial allows hardware builders to publish sensor streams and accept motor commands directly from ROS 2 graphs. This simplifies embedded sensor-actuator bring-up for open-hardware robotics projects.
The maintainer highlights that deterministic C99 memory bounds prevent runtime heap fragmentation, making the library suitable for low-level motor drives. Open-source developers point out that serial transport protocols still impose bandwidth limits compared to native CAN-bus or Ethernet ROS 2 implementations when handling high-frequency IMU arrays.
On Wednesday, November 18, researchers published Z-sight, an embodied foundation model architecture that unifies vision-language-action (VLA) processing with generative latent world modeling. Rather than reconstructing full pixel-space video predictions, Z-sight operates entirely within the visual token space of its underlying vision-language model, using a mixture-of-transformers to denoise future spatial tokens and continuous action chunks. Evaluated on the Unitree G1, the model achieved a 73.3% success rate across six complex manipulation tasks with a control latency of 110 milliseconds per action chunk.
Why it matters
Traditional video world action models incur massive computational overhead by rendering future pixel frames, introducing control latencies that make real-time physical interaction difficult. Z-sight bypasses pixel rendering by predicting future world states directly inside the latent token space used for decision-making. This unified representation enables robots to evaluate future physical trajectories and execute continuous motor commands within a single, low-latency forward pass.
The paper's authors demonstrate that latent-space foresight eliminates the compute bottleneck of visual decoding without sacrificing physical manipulation accuracy. Autonomous policy researchers note that while latent prediction speeds up inference, visualizing and debugging policy failure modes becomes more difficult when pixel-level visual outputs are absent.
On Wednesday, November 18, NVIDIA released SOMA Retargeter v0.2, an open-source motion conversion toolkit built on NVIDIA Warp and the Newton IK solver. The v0.2 update provides pre-configured kinematic mapping files and reference motion libraries for five humanoid platforms: Unitree G1, Unitree H2, Booster T1, AgiBot X2 Ultra, and AgiBot A3T3. The toolkit converts optical motion capture and BVH files into CSV joint-trajectory datasets through automated limb scaling, foot-plant contact correction, and headless batch processing.
Why it matters
A primary bottleneck in humanoid reinforcement learning is retargeting human motion capture data onto robot kinematic structures with varying limb proportions, mass distributions, and joint limits. SOMA Retargeter v0.2 automates this inverse kinematics translation, allowing engineering teams to convert public human video datasets into physics-validated training trajectories. Providing open configurations for leading Chinese and global humanoids lowers the data preparation overhead for sim-to-real training.
NVIDIA presents SOMA as a standardized pipeline to convert raw human motion capture into reinforcement learning datasets across diverse humanoid frames. Independent simulation engineers observe that while automatic foot-plant correction fixes ground penetration, physical torque constraints and actuator thermal limits must still be enforced during downstream physics simulation.
On Wednesday, November 18, Addionics published validation data from Avrion Battery Labs demonstrating that cylindrical 18650 lithium-ion cells incorporating 3D porous current collectors achieved 3.5 times the cycle life of standard foil-based cells. The open porous copper structure enhances electrolyte ion mobility, allowing depleted cells to recover internal salt homogeneity in under 12 hours after 200 heavy discharge cycles, compared to over 48 hours for reference cells at cycle 20. The battery design targets continuous-operation hardware including mobile robots and electric vehicles.
Why it matters
Continuous operational cycles in commercial robot fleets cause severe battery capacity fade and long rest-period delays due to electrolyte motion-induced salt inhomogeneity (EMSI) and lithium plating. By shortening ion diffusion distances through a through-plane porous architecture, Addionics mitigates internal degradation during fast charging and heavy drain. Extending battery lifespan and reducing required rest times directly lowers total cost of ownership for autonomous fleets.
Addionics and Avrion Battery Labs state that 3D current collectors solve fundamental chemical degradation channels without requiring novel, unproven chemistry formulations. Battery manufacturing analysts point out that scaling porous current collectors requires retrofitting existing roll-to-roll electrode coating lines, which may introduce initial manufacturing yield challenges.
On Thursday, November 19, MegaRobo Technologies closed a funding round of nearly $200 million featuring participation from the Oman Investment Authority, Asia Investment Capital, Granite Asia, and Sinovation Ventures. The fresh capital brings MegaRobo's cumulative funding to approximately $600 million. The company plans to allocate the capital toward expanding its global AI for Science business and scaling its One Intelligence Stack—a unified software and hardware architecture for automated laboratory robotics and high-throughput physical experimentation.
Why it matters
Automating pharmaceutical research and biotechnology laboratories requires high-precision robotic manipulation operating alongside complex scheduling software. MegaRobo's capital raise highlights substantial global venture interest in software-defined lab automation platforms that close the loop between digital model predictions and physical chemical synthesis. Unifying hardware cells under a standardized intelligence stack addresses the integration friction that typically slows laboratory automation.
MegaRobo's executive leadership states that a unified intelligence architecture allows life science clients to scale automated experimentation without custom hardware re-engineering. Industry observers note that expanding lab robotics globally requires navigating strict healthcare compliance standards and complex legacy assay equipment integrations.
On Thursday, November 19, PhD researchers at the Bristol Robotics Laboratory (BRL) detailed two assistive robotics initiatives aimed at eldercare and cognitive support. The Socially Assistive Walker (SAW) integrates motorized mobility assistance, micro-prompting language models, and automated braking to support independent walking. In parallel, researchers partnered with Bristol startup Hijimbo to test a domestic assistant designed to help individuals with dementia locate misplaced household items and navigate daily tasks.
Why it matters
Integrating lightweight language models and automated physical resistance into assistive mobility devices bridges the gap between passive medical equipment and intelligent companion robotics. Addressing age-related mobility decline and memory loss through user-aware physical prompting enables safer independent living. These deployments provide blueprints for certified assistive devices entering eldercare environments.
BRL researchers emphasize that co-designing physical mobility hardware alongside conversational prompting improves patient adherence and safety. Clinical care specialists note that assistive devices using LLMs must maintain high reliability and strict offline privacy controls when deployed in eldercare settings.
On Tuesday, November 17, SiMa.ai announced Palette 3.0, a software architecture update designed to run multiple concurrent machine learning models on a single system-on-chip for autonomous robotics and drones. Built for the company's Modalix MLSoC, the software incorporates speculative decoding frameworks—including EAGLE3 and DFlash—to boost token generation throughput by up to 3.5 times. The update expands support for mixture-of-experts architectures such as GPT-OSS-20B on 16 GB system-on-module form factors.
Why it matters
Autonomous edge platforms frequently require simultaneous execution of distinct neural networks—such as vision-language models for high-level reasoning alongside real-time depth estimation and obstacle avoidance. Running these workloads on legacy edge silicon often forces sequential execution or offloading to high-power cloud servers. SiMa.ai's multi-model scheduling layer allows robotics teams to combine perception and reasoning pipelines on local 15-watt edge compute modules.
SiMa.ai highlights that running concurrent models on edge SoCs resolves thermal and power constraints in fanless robotic systems. System integrators note that achieving the claimed 3.5x throughput gain requires precise quantization and memory partitioning, which may vary depending on the specific multi-model topology.
On Thursday, November 19, MYIR launched the MYC-JR3588, a compact Edge AI System-on-Module conforming to the SMARC 2.2 specification. Measuring 82 mm x 50 mm, the board is powered by the Rockchip RK3588 processor, integrating an octa-core Arm CPU, three Cortex-M0 real-time cores, an Arm Mali-G610 GPU, and an embedded 6 TOPS INT8 NPU. Designed for autonomous mobile robots (AMRs) and industrial automation, the module includes 8GB LPDDR5, 64GB eMMC, multiple PCIe 3.0 interfaces, dual Gigabit Ethernet, Wi-Fi 6, and a TPM 2.0 hardware security chip.
Why it matters
Standardized form factors like SMARC 2.2 allow robotics OEMs to upgrade compute cores without redesigning custom carrier boards and sensor pinouts. Combining high-level Linux processing with dedicated Cortex-M0 real-time cores on a single module enables simultaneous execution of ROS 2 navigation stacks and low-latency motor control loops. This reduces board count and power consumption in mobile industrial manipulators.
MYIR positions the MYC-JR3588 as a cost-effective, long-lifecycle alternative to discrete edge GPU boards for high-volume AMR manufacturing. Embedded engineers point out that while 6 TOPS is sufficient for 2D/3D LiDAR processing and classic vision models, heavy vision-language-action policies still require higher-class edge accelerators.
On Thursday, November 19, automotive semiconductor designer Black Sesame Technologies established its embodied AI headquarters in Qianhai, Shenzhen, operating under a dual-headquarters model alongside Hong Kong. The company showcased its SesameX robotics computing platform, spanning the Kalos, Aura, and Liora chip lines designed for wheeled, quadruped, and humanoid robotics. Black Sesame confirmed securing mass-production design wins from several robotics manufacturers, alongside ecosystem partnerships with Midea, AUBO Robotics, and COSCO Shipping.
Why it matters
Automotive chipmakers are increasingly re-architecting their high-throughput SoC designs to capture the emerging physical AI and robotics market. By establishing a dedicated robotics computing platform in Shenzhen's manufacturing ecosystem, Black Sesame aims to supply localized, high-reliability edge silicon for real-time joint control and vision processing. Cross-industry hardware transfer from automotive supply chains accelerates the availability of automotive-grade silicon for commercial humanoids.
Black Sesame's executive team emphasizes that leveraging automotive-certified SoC designs provides higher reliability and lower component costs for humanoid builders. Industry analysts note that established edge compute incumbents like NVIDIA and Qualcomm maintain strong software ecosystem moats that new entrants must match.
Building on the Shenzhen overnight logistics deployments we tracked earlier this week, Chinese robotics firm ROBOTERA presented new operational data at the APEC CEO Summit. CEO Jianyu Chen confirmed the company's M7 humanoid is actively deployed across more than 10 postal hubs operated by China Post and SF Express, achieving induction rates exceeding 1,500 parcels per hour during continuous overnight picking shifts.
Why it matters
Reaching 1,500 parcels per hour matches human shift speeds in parcel induction, offering a measurable benchmark for operators evaluating labor automation. Sustaining this throughput across multi-hub deployments provides concrete operational data validating whole-body motion planning under high-frequency, repetitive picking cycles.
While ROBOTERA positions these metrics as a direct response to the regional overnight labor shortages we've tracked, warehouse automation specialists caution that long-term ROI will depend heavily on mean time between failures (MTBF) and joint maintenance overhead during continuous multi-shift operations.
Verified across 2 sources:
PR Newswire(Nov 19) · CNW(Nov 19)
Click Copy for AI above, then paste the prompt
into your favorite AI chatbot — ChatGPT, Claude, Gemini, or
Perplexity all work well.
In research published November 18 in Science Robotics, MIT engineers led by Kevin Chen demonstrated an insect-scale flapping-wing microrobot capable of autonomous flight using onboard sensors and computation. By increasing the geometric scale of its soft artificial muscles by 50%, the team boosted net lift by 46% to reach a 1.9 lift-to-weight ratio. This extra payload capacity allowed the vehicle to carry a 244-milligram custom electronics package incorporating an inertial measurement unit, an optical flow sensor, and a time-of-flight range sensor to maintain flight stability within centimeters without external motion-capture cameras.
Why it matters
Insect-scale aerial robots have traditionally been tethered to external power supplies or high-speed motion-capture camera arrays for attitude estimation and position control. Packing state estimation and control loops onto a sub-gram vehicle establishes a baseline for deploying miniature swarms in GPS-denied or physically constrained environments like agricultural canopies and collapsed infrastructure. The demonstration proves that lightweight optical flow and IMU algorithms can maintain stability despite high-frequency wing vibrations.
The MIT team emphasizes that achieving sensory and compute autonomy moves sub-gram flight from laboratory demonstrations toward practical field operations. External aerial robotics researchers note that while compute autonomy is solved, flight times remain severely constrained until high-energy-density micro-batteries can be integrated without exceeding lift thresholds.
On Wednesday, November 18, researchers at the University of California San Diego introduced a 3D-printed, six-legged soft robot that achieves locomotion speeds up to 4.72 meters per second using pressurized air without onboard electronic components. Published in Science Robotics, the device is printed in a single piece from a single material and incorporates a double-expansion pneumatic oscillator inspired by compound steam engines. This dual-stage mechanism routes high-pressure gas across chambers twice, boosting energy efficiency over 100-fold compared to prior soft pneumatic designs and generating up to 169 watts of mechanical power.
Why it matters
Integrating multi-stage thermodynamic pressure expansion directly into compliant, single-material structures overcomes the low force-density and slow response times that have historically plagued soft robotics. By completely omitting silicon microcontrollers and copper wiring, the platform can operate in high-radiation, strong electromagnetic, or submerged environments where electronics short-circuit or degrade. The one-shot additive manufacturing process also eliminates manual assembly steps for fluidic logic gates.
Lead researcher Yichen Zhai and Professor Michael T. Tolley emphasize that physical fluidic routing can match the high-speed gait generation of electronic microcontrollers while remaining environmentally resilient. Skeptics point out that the current iteration relies on an external compressed air tether, noted by the authors as a target for future miniaturized, onboard gas cartridge integrations.
On Wednesday, November 18, researchers at KAIST, led by Professor Il-Kwon Oh, detailed a flexible lithium-sulfur battery electrode fabricated by growing nickel cobalt selenide (NiCo2Se4) nanowires directly onto woven carbon fibers. Published in Advanced Composites and Hybrid Materials, the structural cathode sustained 1,000 bending cycles at a 2.5 millimeter radius without structural degradation. When integrated into curved drone-leg pouch cells, the battery retained 99.8% of its initial voltage even after external mechanical loads fractured the surrounding composite frame.
Why it matters
Separating power storage from structural load-bearing frames creates dead weight in mobile robotics and aerial drones. Monolithically growing active nanowire materials on flexible carbon cloth prevents the delamination and polysulfide migration that normally degrade lithium-sulfur batteries under bending stress. This allows structural components like robotic limbs and landing gear to double as energy storage reservoirs, extending mission runtimes.
The KAIST team notes that structural lithium-sulfur batteries maximize spatial mass efficiency in weight-sensitive aerial and legged robots. Energy storage researchers caution that while cathode mechanical flexibility is demonstrated, scaling pouch-cell manufacturing and ensuring long-term moisture sealing in open robotic frames remain key engineering hurdles.
On Wednesday, November 18, Pony.ai reported its third-quarter 2026 financial results, showing a 75.2% year-over-year revenue increase to $44.6 million, driven by $17.8 million in robotaxi services and $18.0 million in robotruck operations. During the earnings call, executives revealed that annual insurance premiums for its 7th-generation robotaxi fleet dropped to 40% of the cost of human-driven taxis due to improved safety records across 2,768 active vehicles. GAAP net losses widened to $136.3 million amid fleet expansion and R&D spending.
Why it matters
Securing a 60% discount on insurance premiums compared to human-operated taxis provides objective third-party validation of autonomous fleet safety performance. Lower insurance costs directly improve unit economics for commercial ride-hailing networks. Concurrently, generating balanced revenue across passenger robotaxis and port robotrucks demonstrates operational leverage across distinct autonomous mobility markets.
Pony.ai management states that lower insurance premiums and expanding contiguous operational zones in Guangzhou prove the commercial scalability of its Gen-7 platform. Financial analysts emphasize that despite revenue growth, narrowing the $136.3 million quarterly loss requires scaling fleet utilization and lowering sensor package costs.
On Tuesday, November 17, startup Neufangled Robotics unveiled Stringman, an overhead cable-driven home robot designed to pick up laundry and objects from floors. Created by founder Nathaniel Nifong, the system utilizes ceiling-mounted motorized cables and a suspended two-finger gripper operating across workspaces up to 5x5 meters with a payload capacity of 2.1 kilograms. The fully assembled hardware is priced at $965, and the company released open CAD files, parts lists, and control software for developer self-assembly.
Why it matters
Mobile floor robots frequently struggle with clutter, threshold navigation, and balance when attempting domestic tidying tasks. Stringman bypasses floor obstacles entirely by employing an overhead cable-driven parallel manipulator (CDPM) to clear objects from above. Open-sourcing the design files at a sub-$1,000 price point allows home automation developers to experiment with alternative kinematics for indoor assistance.
Neufangled Robotics highlights that overhead cable suspension avoids floor clutter and reduces mechanical complexity compared to mobile manipulators. Consumer hardware critics point out that requiring ceiling anchor installations and an external computer for AI processing limits mass-market consumer appeal.
In detailed updates published November 18 from its European Partner Summit, Pudu Robotics outlined its commercial expansion strategy built around 'One Brain, Multiple Embodiments.' The platform decouples shared intelligence layers—powered by foundation model PuduFM and agent framework PuduAgent—from diverse hardware forms including delivery units, quadrupeds, and the PUDU D series humanoids. Pudu confirmed cumulative global shipments exceeding 130,000 units across 85 countries following the opening of its European headquarters in Neuss, Germany.
Why it matters
Deploying general-purpose embodied AI requires re-using foundational navigation and reasoning stacks across varied mechanical form factors. Decoupling the central software intelligence from specific wheel, leg, or arm configurations lowers development costs when introducing new hardware form factors to commercial clients. Pudu's massive operational fleet provides a real-world telemetry baseline for refining embodied foundation models.
Pudu CEO Felix Zhang maintains that field experience across 130,000 commercial delivery robots provides the necessary data foundation to transition into general-purpose humanoids. Industry analysts note that commercial delivery environments differ significantly from complex, contact-rich manipulation tasks, requiring new sensor modalities like force-tactile feedback.
Cleanrooms and Fabs Emerge as Hard-Tech Humanoid Proving Grounds Industrial deployments are expanding beyond automotive assembly and logistics into zero-downtime semiconductor cleanrooms. Partnerships like Boston Dynamics and GlobalFoundries demonstrate that humanoids are being evaluated for sub-fab maintenance and wafer handling, where strict contamination standards and high capital equipment costs impose rigorous operational constraints.
Control-Centric Latent Representations Bypass Pixel Generation Overhead Embodied AI foundation models are abandoning heavy pixel-space video prediction in favor of compact, geometry-aware latent targets. Architectures like Z-sight and LaST1.0 predict future states directly within VLM token spaces or wrist-view latent representations, significantly lowering inference latency and improving real-time closed-loop control.
Electronics-Free Fluidic Logic Pushes Soft Robotics to Extreme Speeds By adapting thermodynamic mechanisms like compound steam engines into 3D-printed pneumatic architectures, soft robots are reaching locomotion speeds of nearly 5 meters per second without microcontrollers or silicon logic. This trend allows compliant machines to operate safely in environments with severe electromagnetic or radiation interference.
Low-Level MCU Integration Connects ROS 2 Directly to Actuator Firmware Open-source communication libraries and modular hardware platforms are embedding ROS 2 nodes directly onto 8-bit and 32-bit microcontrollers without operating system overhead. Projects like ros2serial enable deterministic, low-latency messaging between high-level task planners and low-level motor controllers.
Open Multimodal Datasets Anchor Bipedal and Whole-Body Motion Scaling The release of massive open-source motion datasets containing millions of trajectories is accelerating whole-body bipedal control. Initiatives like Movens-5M demonstrate that scaling motion expert parameters improves trajectory tracking and multi-modal instruction alignment across physical humanoids.
What to Expect
2026-11-25—XPeng AI Day 2026 in Guangzhou featuring live interactions with Iron humanoid robot
2026-12-01—Humanoids Summit 2026 convenes in San Mateo, California
2026-12-05—ROSCon Türkiye inaugural event opens for open-source robotics developers
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
602
📖
Read in full
Every article opened, read, and evaluated
165
⭐
Published today
Ranked by importance and verified across sources
20
— The Robot Beat
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste