Robbyant open-sourced LingBot-Touch, a tactile model that unifies signals from magnetic, visuotactile, and piezoresistive sensors into shared tactile tokens. This model was pretrained on nearly five thousand hours of touch recordings from human operators. Integrating these tokens into robot policies significantly increased success rates across various real-robot tasks compared to using raw sensor data. The research demonstrated that the tactile tokens provided substantial gains for models like LingBot-VLA and T-Rex. Limitations include testing only two hands and a focus on fingertip contact, as well as the exclusion of capacitive sensors from the pretraining data.
Axon invested $1.2 billion in drone maker Skydio and secured exclusive public safety distribution rights in the U.S. and Canada. This investment increased Axon's stake in Skydio to 37%. The deal, agreed upon on November 5, 2026, valued Skydio at $5.2 billion prior to the investment. Axon's payment consisted of 60% cash and 40% stock, funded partly by $1.15 billion in convertible notes. Skydio will remain independent, focusing on autonomous flight while Axon gains a significant sales channel.
NVIDIA and partners at CoRL demonstrated methods for training robots to perform delicate manipulation tasks through simulation. World action models like GR00T 2 enable robots to learn dexterous skills by leveraging diverse data and skill transfer. The Isaac Lab framework, integrating accurate physics, allows for scalable training in simulation using agents and tools like Newton. Agentic robotics, such as the Graph-as-Policy framework, helps coordinate multiple skills into complex robot programs. These advancements, supported by benchmarks like DexBench, aim to bridge the gap between simulation and real-world industrial deployment.
NVIDIA Holoscan 5 enables low-latency robotic control and high-throughput perception through a unified programming model for CPU and GPU workloads. A gravity-compensation loop was demonstrated using three custom Holoscan operators to calculate inverse dynamics with Pinocchio and command a reBot arm. The framework uses execution partitions and drop-oldest queues to minimize latency and jitter in the control path. Profiling with Nsight Systems showed that pinning the control partition significantly reduced p99.99 latency for both C++ and Python implementations. This approach forms a foundation for building more complex systems like hand-guided teaching and manipulation.
NVIDIA introduced GR00T 2, an updated open frontier world action model, and Isaac Lab 3.0, the latest open-source robot learning framework, at CoRL. GR00T 2 enables robots to learn dexterous manipulation skills from diverse data and transfer skills between bodies. Isaac Lab 3.0, integrated with Newton, allows for efficient, parallel simulation training using accurate physics for complex robotic systems. New benchmarks like DexBench and VABAR are emerging to measure robot dexterity across industrial tasks. Various partners are already leveraging these technologies to advance robotics research and industrial applications.
Boston Dynamics attached wheels to its Atlas humanoid to test whole-body control, merging physics-based motor control with reinforcement learning. This experiment aimed to improve dynamic stability across unstructured terrain by allowing Atlas to process real-time feedback. Atlas demonstrated the ability to switch smoothly between rolling on flat surfaces and articulated legged motion over obstacles. The process involved a simple retrofit of an off-the-shelf scooter rim onto the ankle actuator. The research emphasizes that combining classical control with whole-body RL yields superior real-world agility.
Autonomous mobile robots require reliable, fast Wi-Fi roaming to maintain real-time connectivity while moving between coverage zones. Traditional Wi-Fi causes interruptions because the main radio must stop traffic to search for new access points, leading to latency and packet loss. A dedicated scanning architecture, like Synaptics Scan Core, separates network discovery from application traffic using a low-power subsystem. This separation allows for faster roaming decisions, lower latency, and improved power efficiency. Implementing dedicated scanning enhances mobile robot performance by enabling continuous network awareness and faster transitions.
AI agents can utilize simulation to solve complex robotics problems using code instead of relying solely on large datasets. An example shows an agentic robotic arm successfully depositing used hypodermic needles into a disposal box by navigating variable subchallenges, like different box shapes. This approach suggests a potential method for addressing the data-scaling problem in robotics. Agentic systems leverage simulation to enable agents to learn and execute complex physical tasks. Therefore, agentic robotics offers a new paradigm for tackling data limitations in the field.
The MACHINA Summit will return on July 7–8, 2027, at the Palais des Congrès de Paris, expanding to two days with expectations for 3,500 attendees and 200 speakers. Confirmed speakers include Wandercraft CEO Matthieu Masselin, Agility Robotics co-founder Jonathan Hurst, and Skild AI co-founder Abhinav Gupta. The event will feature a closing gala and other related activities like the MACHINA Factory. The summit aims to create a European meeting place for the physical AI industry.
TC-PEBRAS is hosting a video competition to educate a general audience about robot performance evaluation and benchmarking. Participants must create 3-minute videos covering topics like metrics, test methods, and reproducibility in robotics and automation. Submissions are due by December 12, 2026, via email to tc.pebras.video@gmail.com. Four $250 awards will be given for general scope and Humanoids 2026 activities. The competition opens on November 5, 2026, with winners announced on December 17, 2026.
The best machinery positioning method depends on meeting four documented requirements: clearance, load, movement, and floor condition. Options include hydraulic toe jacks, low-profile machinery sliders, wheeled skates, air casters, overhead cranes, and forklifts. Buyers should evaluate positioning equipment against these four criteria for each move rather than defaulting to available tools. Machinery sliders offer a low-profile solution for tight spaces and can be used with multidirectional styles for maneuvering. A comprehensive move checklist should include weight, clearance, floor capacity, debris removal, and alignment plans.
GenRobot AI raised $100 million in a Series A+ funding round to expand its robot training data infrastructure. The company focuses on the equipment and software needed to capture, process, and organize human demonstrations for robot training. This infrastructure addresses the challenge of obtaining sufficient high-quality data for teaching robots complex physical tasks. GenRobot AI provides recording systems and a software platform called Matrix to prepare raw data for machine learning. The investment reflects growing investor interest in the foundational data infrastructure supporting embodied artificial intelligence development.
Luck Stone and Caterpillar are expanding autonomous hauling technology to two additional Virginia operations, building on success at Bull Run Quarry. This expansion leverages autonomous trucks that have hauled over 3.5 million tons since implementation. The collaboration focuses on improving safety, production consistency, and associate skill development through technology. The expansion involves deploying Caterpillar's autonomous hauling solution on Cat 775 trucks at the Boscobel and Bealeton sites. Collectively, Caterpillar autonomous trucks have achieved significant milestones in hauling and safe operation.
Workers must be kept informed about changes in automated factory layouts, robot cell locations, and autonomous mobile robot routes to ensure safe and efficient operations. Automation causes more frequent changes than traditional lines, necessitating a faster method of communication than paper notices. Digital signage offers a flexible way to display dynamic, high-contrast information like production targets and safety messages at the point of work. These displays should be integrated with live automation data, such as output counts, to provide accurate, relevant updates. Ultimately, screens help the human element of the factory keep pace with the rapid evolution of machinery.
TLD Robotics was formed in September 2026 by combining the expertise of EasyMile, TractEasy, and TLD's existing vehicle activities. This consolidation brings together technology, manufacturing, and operational experience developed since 2017. The company aims to simplify the integration of autonomous driving, vehicle manufacturing, and fleet management for ground operations. This integration allows TLD Robotics to offer a complete system, including software and physical platforms, to customers. The new entity seeks to address the broader challenges of deploying autonomous vehicles into routine industrial and airport environments reliably over the long term.
Nature Will launched Natura-1, a 50-billion-parameter physical general foundation model for dexterous hand control. This model was pre-trained on over a million hours of egocentric human data to combine instruction following with the ability to manipulate under changing conditions. Natura-1 learns to connect human intent to physical action through multimodal training on language, images, video, and actions. It operates with a 100-millisecond control loop to unify language, vision, and high-degree-of-freedom action. This work represents a step toward general foundation models capable of interacting effectively with the real world through physical manipulation.
Berkshire Grey has introduced new AI enhancements for its robotic automation portfolio using the BG Pilot software. These upgrades include advanced perception for trailer unloading, intelligent orchestration for store replenishment, and AI-driven precision packing. The BG Pilot software leverages insights from extensive real-world warehouse operations to manage variability effectively. It powers specific product features: Scoop for perception, Stride for orchestration, and Core for packing. This innovation supports growing demand for robotic automation by enabling robots to handle complex, changing warehouse workflows with greater reliability.
Cyngn launched DriveMod EasyLaunch, a subscription service to facilitate the deployment of autonomous material handling for manufacturers and warehouses. This program allows qualifying customers to automate repetitive material movement routes using Cyngn's autonomous tuggers on a monthly basis. EasyLaunch requires no upfront capital and can be deployed in weeks, removing implementation friction. The service focuses on indoor movement of items like carts and pallets between various operational areas. Cyngn manages vehicle deployment and configuration, with customers monitoring activity via Cyngn Insight software.
Comau and Automha showcased integrated food and beverage automation solutions at Cibus Tec 2026, combining robotics, automated storage, and intralogistics technologies. Comau introduced new food-grade robots and wearable devices designed for food handling and ergonomic support. Automha displayed automated warehouse technologies, such as multi-directional shuttles, to improve storage density and material handling. The joint presentation emphasized the need for manufacturers to connect production with automated logistics systems. This integration aims to provide a single automation partner for optimizing the entire food and beverage value chain.
Supervised autonomy requires designing for the robot's worst day, focusing on handling unforeseen situations rather than just smooth operation. Exceptions must be inventoried, with some resolved autonomously and others always requiring human intervention, especially for safety or customer rules. Requests for help must include context, confidence estimates, and time constraints to effectively involve supervisors. A clear supervisor role is necessary, balancing robot autonomy with human oversight through data-driven ratios and new skills. Documenting these limits protects operators commercially and addresses concerns regarding trust in autonomous systems.
ANYbotics has started delivering its ANYmal X quadruped robot for use in hazardous industrial environments like oil, gas, and chemical facilities. This legged inspection system autonomously examines assets by using thermal cameras, acoustic sensors, and gas detectors to find potential issues. The robot is engineered to meet ATEX and IECEx explosion-protection standards through features like nitrogen-filled enclosures and flameproof protection for its components. ANYmal X operates with Shift software to integrate inspection data directly with industrial maintenance systems such as SAP. Early customers include Equinor and DSM-Firmenich, and major energy companies like BP and Shell have participated in testing.
Honda launched a $10,000 competition in partnership with Base 11 and Freelancer to encourage young Americans to propose ways to improve human-robot collaboration in manufacturing. The competition, called the Automation for Good: The Advanced Manufacturing Challenge, invited US residents aged 18 to 26 to address workplace hazards using robotics. Participants were encouraged to propose solutions for tasks involving repetitive lifting, hazardous environments, and ergonomic risks. The initiative aims to introduce young people to advanced manufacturing careers by focusing on automation that supports workers. The competition offers monetary prizes and mentoring opportunities for finalists.
Rogue Cortex, a UK robotics startup, has secured $4.5 million in funding to simplify robotic system development through software and hardware tools. The company launched a hardware-agnostic Software Development Kit (SDK) enabling development without hardware constraints. They also introduced the Drone Developer Kit (DDK), a configurable drone platform with integrated avionics. This approach aims to let developers focus on application creation rather than hardware integration. The company offers its tools via a per-system licensing model.
A Controls Engineer used the Atlas robot to test the limits of its actuators by retrofitting wheels onto the robot's feet. This experiment aimed to see how a learned whole-body controller could work with real, load-bearing hardware, pushing the motors to their performance envelopes. The process involved creating motion trajectories from human data and training a reinforcement learning policy to control the rolling motion. The work highlighted the necessity of closing the simulation-to-real gap by ensuring simulated motor behavior accurately reflected the physical hardware. Ultimately, this project demonstrated the synergistic relationship between low-level motor control and high-level reinforcement learning.
A robotic car charger uses suction cups to firmly attach to the vehicle while charging. The robot utilizes API calls to remotely unlock and open the charging port, bypassing potential wireless attacks. It uses Mecanum wheels and LiDAR for positioning, and an onboard camera with ChARuco patterns for fine-tuning alignment. Suction cups and a gear assembly are used to apply the necessary force to plug in the charger. The robot can also retract by swiveling its camera to see wall markers for parking.
Pilatus Aircraft is testing Hexagon Robotics’ AEON humanoid to perform quality checks on the airframe of its PC-24 business jet in Stans, Switzerland. The testing focuses on quality control tasks within the manufacturer's production operations rather than aircraft assembly. The AEON platform integrates sensors, AI, precision measurement, and motion control to autonomously inspect parts and navigate environments. Pilatus aims to use this technology to improve quality inspections and assist workers with repetitive or demanding tasks. Future plans involve deploying humanoids to enhance safety, autonomy, and employee assistance.
Avatar is a wheeled humanoid prototype from Honda R&D Co., Ltd. designed for remote physical work, featuring two arms with five-fingered hands. It measures 155 cm tall, weighs 101 kg, and has a limited skill score of 3/10, placing it in the developing range. The system utilizes tactile feedback and visual cameras to allow operators to remotely guide the robot in tasks like sorting and assembly. Operator interaction is facilitated via a VR headset and gloves, with AI assisting in executing fine motions. Honda intends to apply Avatar in areas such as medical support, equipment maintenance, and disaster response.
UM1 Evo is a biomechanical robotic hand developed by Didier Vançon. It scored 6 out of 10 overall, highlighting its dexterity over raw strength, with a perfect 5/5 rating for Degrees of Freedom and a 1/5 for Strength. The hand weighs 1.7 kilograms and features 25 electric actuators assembled from custom CNC machined parts. Vançon spent five and a half years developing the prototype by integrating biomechanics, manufacturing, electronics, and software. The project aims to commercialize the prototype and knowledge for applications like prosthetics and humanoid robotics.
Anigon is an industrial wheeled humanoid developed by Aimbo Robotics for logistics, manufacturing, and warehouse environments. It stands 189 cm tall, features force-controlled arms with 28 degrees of freedom, and includes tactile sensing and RGB depth cameras for object interaction. Supported capabilities include obstacle avoidance, sorting goods, and product assembly, utilizing its mobile base and manipulation hardware. The robot is equipped with hot-swappable batteries and safety features like collision detection and electromagnetic brakes. Anigon is currently a prototype with enterprise software integration capabilities, though its skill scores indicate development in navigation and manipulation.
X Hand M1 is a rigid robotic hand manufactured by Yuequan Bionics in Beijing, China, designed for industrial applications following human hand proportions. It features five fingers, measures 190 x 90 x 35 mm, and weighs 0.6 kg. The hand received a Humanoid.Guide Hand Score of 5 out of 10, indicating it is a developing option rather than a top-rated product. Component ratings suggest moderate degrees of freedom and lower strength, emphasizing the need for task-specific evaluation. The product is listed as in production with an advertised price of USD 7,000.
FF Nova is a compact humanoid robot from Faraday Future Intelligent Electric Inc., priced at $2,490, designed for developers, educators, and researchers. It features 20 degrees of freedom, electric servo actuators, and sensing capabilities including a camera, ToF sensor, and 2D LiDAR. Demonstrated skills include obstacle avoidance and standing up after a fall, positioning it as an introductory educational platform. The robot has limited navigation and manipulation skills, indicating it is a starting point for exploration rather than a mature system. It supports development via an open SDK and provides hardware for studying environmental perception and balance recovery.
Star is a 172 cm tall, 72 kg bipedal humanoid robot developed by TOPSTAR Group for industrial manufacturing. It features seven-axis arms with a dual arm payload capacity of 10 kg and 32 degrees of freedom. The robot can walk at a speed of 3 km/h and has demonstrated capabilities in obstacle avoidance and sorting goods. Currently, its skill score is low, with navigation and manipulation rated at preliminary levels. Star is positioned as an early-stage prototype exploring a humanoid format for factory automation.
KAIST is a 165 cm tall, 75 kg humanoid research prototype developed by the Dynamic Robot Control and Design Lab. It demonstrates capabilities in obstacle avoidance, running 100 m, walking stairs, and jumping and landing. The robot's Humanoid Guide Skill Score is 4/10, with navigation rated 4/5 and manipulation rated 0/5, highlighting its strength in mobility. It utilizes internally developed components and reinforcement learning for its control system. The prototype is intended primarily for research, with a listed price of USD 150,000.
Waymo has launched the Accessibility Network UK to ensure autonomous mobility is inclusive and aligns with local needs in the UK. This network facilitates a two-way dialogue between Waymo and disability advocates to shape technology deployment. The network provides feedback to inform features like vehicle-finding tools, audio cues, and screen reader compatibility. It details specific accessible features, including sound alerts, visual identifiers, and customizable in-vehicle options for passengers. This initiative actively involves organizations like the RNIB and Guide Dogs UK in developing a more accessible autonomous travel experience.