Tsing Standard released the automotive-grade first-person data collection solution Tsing-Ego PRO during SAECCE 2026. It uses 8-camera vision capture across the head and both wrists, supports 1080P global exposure and hardware synchronization, and achieves end-to-end synchronization accuracy of ≤1ms.
At the October 29, 2026 AgiBot Innovation Day, AgiBot proposed for the first time the Worken (effective intelligent work hours) metric for measuring a robot's effective workload, and released its third-generation hardware architecture AHA3, the heavy-duty embodied robot Genie G2 Max, and the embodied intelligence operating system Lingqu OS 1.0.
After its first field test, the DIY unmanned surface vehicle BN-USV has entered the V2 design phase, with the focus shifting to a replaceable segmented hull, a CAN distributed control architecture with a main module plus two propulsion modules, and priority on reliable manual control rather than autonomous navigation. In V2, a Raspberry Pi 5 handles high-level processing, while ESP32 controllers handle low-level control and safety, requiring propulsion modules to respond safely and independently when communication fails. V2 is still in development; the above are design goals rather than verified improvements, and the long-term goal is to become a modular platform for marine robotics experimentation and teaching.
At its Innovation Day, AgiBot released the GO-3 Preview embodied foundation model, the WITA-OMNI native multimodal interaction model, the AGILE general motion intelligence algorithm framework, the third-generation hardware architecture AHA3, and Lingqu OS 1.0, covering the three intelligence lines of operation, interaction, and motion.
Under its Gravity embodied intelligence framework, Geek+ has launched the Gravity Control causal intervention model, which can correct local action deviations in robots with a single natural-language prompt, without updating the original policy weights or adding new demonstration data.
Reachy Mini released its "Level UP" update, adding a hoverboard to improve mobility and providing iOS and Android mobile apps so users can interact with the robot without a computer. The update also adds NFC tags, enabling personalization such as hats corresponding to different personalities. In the demo, the hoverboard touched the ground due to an IMU setup error, which the publisher said was an operational issue.
Memory demand for humanoid robots has become a hot topic, sparked by Musk's claim that humanoid robots' RAM needs will surpass those of every other computing device on Earth.
Pollen Robotics has introduced three new features for Reachy Mini: a wheeled Hoverboard chassis that lets the robot move around and explore, Android and iOS mobile apps for setup and control directly from a phone, and NFC hats for switching personalities and unlocking new interactions and apps. The company says these updates offer more freedom and creative space for exploring physical AI.
Topstar has launched a humanoid dual-arm robot solution for sorting irregular parts. Operators issue commands in natural language, and TOP Agent parses the tasks and coordinates visual perception, motion control, and force-controlled gripping, cutting changeover time from hours to minutes. On the hardware side, the humanoid dual arms have a rated payload of 8 kg and repeat positioning accuracy of ±0.05 mm, are equipped with a binocular vision system and a force-controlled adaptive gripping module, and support 24/7 continuous operation. The solution has completed deployment validation in the supply chain of a leading 3C brand and can optionally be equipped with edge computing and industrial large models to extend to processes such as appearance inspection.
Intel will hold an Embodied Intelligence and Robotics Frontier Trends Exchange on November 10 in Zhangjiang, Shanghai, limited to 100 attendees, showcasing a multi-channel GMSL real-time perception solution based on the Intel Core Ultra X7 358H platform and a 358H embodied intelligence complete system solution.
The Yufeng 250-L LiDAR Edition quadcopter drone kit is aimed at research and education. It consists of three core modules: a flight controller, an onboard computer, and LiDAR. It can deploy SLAM, planning, and control algorithms to achieve real-time navigation in unknown environments.
Qiyuan robots have been integrated with Tencent WorkBuddy. Users issue commands in natural language; WorkBuddy understands the intent and breaks down the task, while Qiyuan robots perceive the environment and execute it, forming an interaction chain from task understanding to robot execution.
RoboCanvas, a visual tool aimed at ROS 2 beginners, has released a free public beta that supports visually designing and configuring robots, validating designs, and generating corresponding ROS 2 project files. The author says the goal is to help students and hobbyists avoid detours with URDF, launch files, dependencies, and simulation environment setup, but the tool will not replace learning ROS 2, and feedback from people with ROS 2 experience is still being sought.
RoboCanvas has released a free public beta that supports visually designing and configuring robots, validating designs, and generating corresponding ROS 2 project files. The author says the goal is to lower the barrier to entry for ROS 2, URDF, launch files, and simulation environments, mainly for students and hobbyists, and stresses that the tool will not replace learning ROS 2.
REEMAN ROBOT partnered with a German industrial technology customer to develop a mobile manipulator robot for warehouse shelf picking and material handling. The solution uses the Big Dog Chassis AMR platform with a rated payload of 100 kg, paired with a 7-DOF robotic arm with a rated payload of 5 kg and a working reach of 610 mm, as well as a vertical lifting mechanism with a stroke of 1,200 mm.
The REEMAN Big Dog wheeled AMR base is designed for mobile manipulator integration, with a rated payload of 100 kg, laser SLAM navigation, LiDAR and dual 3D cameras, and HTTP API, Android SDK, RS232, and local MQTT interfaces. Two cases show a customer integrating dual arms on the Big Dog base for a tote-handling demo, and a German industrial customer combining the base with a 7-DOF robotic arm and a 1,200 mm vertical lift mechanism for warehouse shelf picking.
The author used a Python script and modified firmware to synchronize the OpenMV AE3 camera's exposure with specific IMU measurements, aiming to improve visual-inertial odometry. The firmware change adds a shared clock between the two CPU cores; the IMU runs at 208 Hz, and every 10 IMU measurements trigger one camera exposure, at about 20 FPS. The code is open-sourced, and test videos of Kimera-VIO in 3 m/s outdoor flight and indoor flight are provided.
Gu Yue Ju published a source-code walkthrough of OpenWAM, covering the project structure, the Dual-System architecture, Video/Action joint attention, the Flow Matching training pipeline, and inference deployment, organized around key code. The material provides only an abstract and does not give specific code details or conclusions.
The author documents the full process of deploying the CLIP vision encoder to the RK3588 NPU, covering model export, conversion to RKNN, and mixed quantization, and collects the problems encountered along the way. The material provides only a summary and does not give specific quantization configurations or performance data.
For the RK3588 platform, this approach uses RKLLM to move task planning from the cloud to the edge, letting the robot's main controller reason about the next action on its own. The original text does not provide a specific model version, task scenario, or measured performance data.
After INT8 quantization on RK3588, accuracy dropped by 6.4 points; the article discusses how to remedy this accuracy loss. It focuses on quantization accuracy issues on this platform and possible approaches to address them.
The article describes using RK3588 with GPT-6 Astra to control an industrial robotic arm via natural language, covering task parsing, tool invocation, action execution, and exception handling, with code examples, and explains the complete pipeline from instruction to execution.
A customer integrated a dual-arm mobile manipulator on the REEMAN Big Dog wheeled AMR chassis; in an 82-second demo, it completed dual-arm grasping of a blue material box, transport between workstations, and placement. The chassis has a rated payload of 100 kg, uses laser SLAM navigation, is equipped with LiDAR and dual 3D cameras for obstacle avoidance, and offers development interfaces including an HTTP API, Android SDK, RS232, and local MQTT. The video is a customer integration demo, not evidence of a production deployment or verified throughput performance.
The Stardust OS (AOS) robot hackathon finals, hosted by Stardust Intelligence, concluded in Shenzhen on October 25. A total of 47 teams took part across three stops; in the finals, 17 teams developed 16 apps over 36 hours, with a 94% rate of running on real robots. AOS packages perception, mobility, manipulation, and interaction into callable Meta packages and Skills, and creators combine large models such as GPT and DeepSeek to complete development, simulation testing, and real-robot validation starting from a one-sentence requirement. The champion team, "Remote Avatar," achieved two-way real-time control between a phone and the T1 and completed tasks including scooping cat litter, watering plants, and picking up packages.
At the 2026 Data Storage Industry Conference, Sugon showcased an "AI Data Factory" for the Token era, promoting storage's active participation in the entire AI production process to improve Token output efficiency. Sugon storage systems have been applied in multiple scenarios including Sugon 8000, robot data collection and training, and autonomous driving.
Quasi Robotics has launched a wireless automatic docking and charging solution for its Model C2 series AMRs, using WiBotic wireless charging technology with up to 300 watts of power, supporting automatic replenishment based on battery level, preset times, or fleet management system commands. The solution is available for the C2 Large, C2 PartPorter, and C2 Heavy platforms; the C2 Mini and Standard are not currently supported. The system transfers energy through paired coils and tolerates a certain positional offset, reducing reliance on mechanical docking precision.
The robot is autonomously controlled by AI in the cloud. At the Long Horizon Summit, @Ryan_Resolution showed me a compelling demo: a MakerMods robotic arm controlled via a public API by Dream Lab's latest world-action model. @makermodsai @dreamlabs_ai
Kardex has released BinInductAI, which uses AI and computer vision to replace manual pre-checks and barcode scanning in the AutoStore bin induction process, cutting on-site induction time by up to 30%. Operators only need to place the bin at the induction port; the system automatically checks that the bin is present, its walls, the robot gripper holes, and barcode readability, and after barcode verification confirms with the ASRS that the bin is not registered as a duplicate before issuing the insert command. The system can automatically identify damaged or non-standard bins, reducing the risk of downstream interruptions.
The creator published a 20-minute full build log of the ESP32-S3 desktop robot Pamk, covering the process from changing the initial sketch to a cube shape and designing a custom PCB costing about 25 euros (integrating ESP32-S3, I2S microphone and amplifier, screen, USB-C) to assembly and 3 magnetic shells. Pamk currently has no wheels or actuators, and the author says V2 plans to add motors/actuators. The video has French voiceover and English subtitles.
NVIDIA RTX Spark laptops add the Nemotron 3.5 Transcribe open-weight speech model, support for Liquid AI's LFM3 series models, and an agent resource management feature in the NVIDIA App that allocates GPU resources.
NVIDIA and Agile Robots launched the Agile ONE SN humanoid robot reference platform at GTC Berlin, integrating Isaac GR00T and Jetson Thor, with plans to release it in early 2027.
The Stardust OS Robotics Hackathon finals concluded in Shenzhen, with 16 of 17 teams' apps moving from simulation to real-robot testing, a real-robot rate of 94%. The competition centered on generating apps from a single sentence; across three stops it drew 47 teams, with entries covering tasks such as follow-cam photography, elderly care, and remote avatars. Stardust Intelligence uses AOS to connect external tools such as GPT-6 and Claude along with its in-house Lumo model, and to dispatch the capabilities of its cable-driven robot body downward.
KUKA Mobile Robotics will exhibit the KMP L series latent lifting mobile robots, a one-stop pallet handling solution, and the KUKA.AMR software 3.0 platform at CeMAT ASIA in Shanghai from November 3-6, 2026; together, the three form the warehousing and logistics implementation of the "KAI" intelligent platform. The booth will also demonstrate a full-scenario, multi-system collaborative operation loop covering receiving and inbound storage, material transfer, high-bay storage, and sorting and outbound.
A hands-on explanation of the Zephyr DMA subsystem, covering controllers, channels, and double-buffered transfers. The original text does not provide verifiable metrics such as specific controller models, channel counts, transfer rates, or experimental conditions.
A practical guide to Zephyr's device tree and device driver model, following the complete path from DT binding and driver registration to device retrieval. It covers the key steps from device tree description to driver registration and runtime device retrieval, aimed at developers who need to understand how Zephyr's driver model is implemented in practice.
The JAKA Kargo wheeled humanoid robot, running the JAKA EVO embodied intelligence operating system, has been deployed for bin handling at the headlight plant and electronics plant of a leading Chinese auto parts manufacturer. Per-bin handling time dropped from 105 seconds per bin to 28 seconds per bin, nearly a 4x efficiency gain, and the robot also stacks bins five layers high and recovers empty bins by layer.
On October 25, Stardust Intelligence held the Stardust OS Robot Hackathon finals in Shenzhen, where 17 teams took 16 apps from simulation to execution on the T1 humanoid robot within 36 hours, a 94% deployment rate. Stardust OS (AOS) packages perception, mobility, manipulation, and interaction into Meta packages and Skills, letting developers orchestrate tasks by describing intent in natural language. The champion team, "Remote Avatar," connected a mobile phone, the Stardust desktop, and the T1 real robot to complete a closed loop of multi-step tasks including scooping cat litter, watering plants, pouring water, and picking up packages.
Robot lamp Lamp announced its October development progress: the main development branch has already seen 4,092 commits this month, surpassing the previous high of 1,615 in April. The project added a docs/develop.md development path that takes about 10 minutes to get started, including a simulated lamp on a laptop, Python and JavaScript SDKs, and lamp-mcp; camera images are now authorized separately per skill, and a yield detection based on servo position and load was added, with this feature off by default.
The 32-to-16 qualifying round of the first URKL Global Humanoid Robot Free Combat League has ended, with 16 teams advancing to the group stage and each receiving an EngineAI T800 robot. The qualifiers held six days of offline training at the Dasha River Cultural and Sports Center in Nanshan, Shenzhen. All teams used the T800 as their unified hardware platform, deploying code refined in simulation onto real robots for the first time, and were evaluated on basic combat moves such as getting up after a fall, navigating complex terrain, and punching and kicking. The event features a prize pool of tens of millions, and industry partner Dison Digital will provide the top 16 with motion capture experience benefits.
Intel will hold an exchange on frontier trends and solutions for embodied intelligence and robotics on November 10, 2026, in Zhangjiang, Shanghai, capped at 100 attendees. The event will feature a live demonstration of a multi-channel GMSL real-time perception solution based on the Intel Core Ultra X7 358H (Panther Lake) platform and a 358H embodied intelligence complete-machine solution.