This week's Automate 2026 trade show in Chicago confirms the robotics industry is moving from pilot projects to production, with a heavy focus on humanoids and the 'physical AI' that powers them. The other major theme is safety, as key players roll out new platforms designed to help autonomous systems work alongside humans.
As we continue tracking the Automate 2026 trade show in Chicago, the focus remains firmly on humanoids and 'physical AI.' Alongside the NVIDIA-sponsored Humanoid Robot Pavilion and systems from Boston Dynamics and Agility Robotics we noted yesterday, the floor features new downstream hardware and software solutions. Vendors like Vecow and DFI are debuting edge AI compute platforms built on NVIDIA and Intel silicon to power these next-generation robots, while companies like InOrbit.AI are demonstrating how to orchestrate multi-vendor fleets in commercial deployments.
Why it matters
The heavy focus on humanoids at Automate reinforces that the technology is transitioning from R&D to serious consideration for factory floors. The supporting ecosystem of edge compute hardware and fleet orchestration tools underscores the market's maturation beyond just the robots themselves.
While organizers push the 'pilot to production' theme, the presence of companies like InOrbit.AI highlights a key downstream problem for real-world deployment: orchestrating fleets of diverse robots from multiple vendors.
Building on the humanoid programs we've been tracking at BYD and XPeng, a broader trend is crystallizing: major Chinese automakers, now including Chery, are pivoting to humanoid robot production. Leveraging existing factory infrastructure, this strategic shift—echoing Hyundai's recent full acquisition of Boston Dynamics—is driven by a contracting car market and the massive potential of embodied AI. These companies increasingly view their core competency as mass-producing complex electro-mechanical systems, rather than just vehicles.
Why it matters
As we've noted with BYD and XPeng's factory-first strategies, the entry of automotive giants into the humanoid race could dramatically accelerate industrialization. Their deep experience in supply chain management and high-volume manufacturing is exactly what the robotics industry needs to move from prototypes to mass-market products.
A new analysis from 'The Digital Speaker' argues that the real asset of these automakers is their production capability, not just their car brands. As vehicles become commoditized, these industrial giants are looking to embodied AI as the next major growth platform.
Jean-Baptiste Kempf, the lead developer of the widely used VLC Media Player, has raised a $5 million seed round for his Paris-based startup, Kyber. The company is building an open-source, real-time infrastructure layer designed to connect human operators and AI agents to physical machines like robots and drones. Kyber's protocol focuses on providing low-latency, synchronized streams of video, audio, sensor data, and control inputs, aiming to become a universal standard for teleoperation.
Why it matters
As robotic fleets scale, the need for a robust, standardized, and low-latency communication and control layer becomes critical. Kempf's strategy of building an open-source protocol, similar to his approach with VLC, could democratize access to high-performance remote operation capabilities, which are currently often proprietary and fragmented. This could accelerate development and interoperability for a wide range of applications, from warehouse logistics to remote industrial inspection and defense.
This is a developing story, following up on Kyber's launch last week. This new report confirms the seed funding amount and investors. Kempf argues that just as the internet needed HTTP, physical AI needs a foundational protocol for real-world interaction. He believes the layer underneath physical AI is an overlooked but massive opportunity, aiming to own the infrastructure before large-scale robot fleets become ubiquitous.
NVIDIA has launched Halos for Robotics, a full-stack functional safety system designed to accelerate the deployment of robots in environments shared with humans. Announced on Monday, the system extends NVIDIA's safety architecture, originally developed for autonomous vehicles, to industrial robots, humanoids, and autonomous mobile robots (AMRs). The platform comprises three main parts: the NVIDIA IGX Thor hardware, the Halos OS software with safety-certified components, and a new Halos AI Systems Inspection Lab to help partners with certification. Agility Robotics is named as an early adopter, integrating Halos into its Digit humanoid robot for industrial applications.
Why it matters
The lack of a standardized, certifiable safety framework has been a major barrier to deploying advanced robots, especially humanoids, at scale in complex, unstructured environments. By productizing safety and leveraging its extensive work in the automotive sector, NVIDIA is providing a critical piece of infrastructure that could significantly de-risk and accelerate commercial rollouts. For robotics entrepreneurs, this offers a clearer, albeit potentially NVIDIA-centric, path to regulatory compliance and enterprise adoption, turning a bespoke engineering challenge into an integrated platform feature.
NVIDIA's announcement frames this as extending its 'proven safety foundation' from the automotive world to the broader robotics market, a move essential for 'bridging the physical AI safety gap.' Early adopter Agility Robotics noted the importance of this for operating in 'human-centric environments' like factories and warehouses. The system is designed to unify AI compute and safety functions that are traditionally separate, a key step for enabling intelligent machines to make safe, real-time decisions.
Naver Labs Europe introduced DUNE on Tuesday, a universal AI encoder that consolidates multiple specialized encoders into a single, efficient model. This innovation significantly reduces the memory and computing power required to run sophisticated AI functions on robots. By creating a unified representation for different tasks and sensor inputs, DUNE allows high-performance AI capabilities to be deployed on a more diverse range of robot hardware with limited resources.
Why it matters
Computational overhead is a major obstacle to commercializing physical AI. Many advanced perception and reasoning models are too resource-intensive for cost-effective edge devices. DUNE addresses this by making robotic intelligence more memory- and compute-efficient, potentially accelerating the adoption of AI-powered robots in both industrial and consumer settings by lowering hardware requirements and costs.
Naver Labs presented DUNE at the Conference on Computer Vision and Pattern Recognition (CVPR), positioning it as a key technology to 'democratize AI for robotics.' The ability to run powerful AI on less powerful hardware could expand the market for intelligent robots beyond high-end, specialized systems.
The University of Oxford will host the new British Open-ended Learning and Discovery Lab (BOLD), a £60 million government-backed initiative aimed at developing new paradigms for artificial intelligence. The lab will focus on human-centered, resource-efficient AI, with a specific emphasis on embodied learning for robots. The goal is to move beyond simply scaling up existing large model architectures and explore more efficient and sustainable learning methods.
Why it matters
BOLD's focus on new learning paradigms, especially for physical robots, represents a strategic investment in the next generation of AI research. By prioritizing resource efficiency and open systems, the initiative aims to counteract the trend of ever-larger, computationally expensive models dominated by a few large tech companies. This could foster a more diverse and sustainable AI ecosystem and position the UK as a leader in developing practical, efficient AI for real-world interaction.
The lab's mission is explicitly not to 'just build bigger versions of what we have now.' Instead, it aims to pioneer new learning approaches and foster open research, training a new generation of AI experts in the UK. The emphasis on embodied AI acknowledges that physical interaction is a critical frontier for artificial intelligence.
Star Dynasty, a humanoid robotics company affiliated with Tsinghua University, has raised over RMB 4 billion (approximately $550 million) and has begun deploying its bipedal robots in SF Express logistics centers across China. The deployment marks one of the first large-scale commercial uses of humanoid robots in the demanding logistics sector, creating a full 'commercial closed loop' for the company's embodied AI technology.
Why it matters
This is a significant proof point for the commercial viability of humanoid robots beyond controlled demos. Securing a massive funding round and immediately deploying units with a major logistics partner like SF Express demonstrates a level of market traction and production readiness that few humanoid companies have achieved. It validates the use case for humanoids in unstructured environments and signals that the Chinese market, spurred by government mandates, is rapidly moving toward real-world adoption.
This move is a tangible result of China's national strategy to lead in embodied intelligence. The combination of top-tier academic research from Tsinghua, substantial capital, and a clear commercial application with a logistics giant creates a powerful model for accelerating robotics development and deployment. The company's success will likely intensify the global race to automate logistics and other labor-intensive industries.
Oversonic Robotics, an Italian company known for its RoBee cognitive humanoid robot, has brought on STMicroelectronics, Fondazione ENEA Tech Biomedical, and SpotInvest as new shareholders. The investment is aimed at accelerating Oversonic's industrial and international growth, particularly in the United States. The company plans to use the capital to expand the application of its humanoid robots in manufacturing and healthcare settings.
Why it matters
The backing of a major semiconductor manufacturer like STMicroelectronics and a biomedical foundation is a strong vote of confidence in Oversonic's technology and its target markets. This strategic investment provides not just capital but also deep technical and domain expertise, potentially strengthening Oversonic's supply chain and accelerating its product development for certified medical and industrial applications. It positions the company as a key European player in the global humanoid market.
Oversonic states the partnership will enhance the 'Made in Italy' technological supply chain and support its expansion into the U.S. The investment from STMicroelectronics suggests a strategic interest in the components and silicon required for cognitive robotics. Oversonic's RoBee robot is designed to operate in social contexts and perform tasks to support human workers, aligning with the growing demand for automation in sectors facing labor shortages.
A benchmark study published in Nature Medicine on Tuesday found that general-purpose large language models (LLMs) like GPT-5.2 and Gemini 3.1 Pro consistently outperform specialized, FDA-cleared clinical AI tools in answering real-world questions from physicians. The study highlights a major regulatory gap, as the FDA's 510(k) clearance process often focuses on the architectural safety of a device rather than its comparative clinical performance against state-of-the-art alternatives.
Why it matters
This study is a bombshell for the medical AI industry. It suggests that expensive, certified AI tools may be less effective than more accessible general models, creating a dilemma for hospitals and clinicians. For developers of medical robotics and AI, it exposes the inadequacy of using FDA clearance as a sole benchmark for quality or utility. The findings will likely force the FDA to re-evaluate its validation requirements, pushing for performance-based evidence and continuous monitoring rather than point-in-time architectural reviews.
The authors argue that the current regulatory framework leaves healthcare providers and clinical trial sponsors without a reliable way to assess an AI's fitness for purpose. The reliance on 'substantial equivalence' to older, less capable devices may be inadvertently stifling innovation and locking in inferior technology. This creates a significant challenge for companies that have invested heavily in the lengthy and expensive FDA clearance process, only to be outperformed by rapidly evolving general models.
The FDA has issued several 510(k) clearances for AI-powered medical technologies this week. Masimo received clearance for an AI feature in its wearable monitor that predictively detects Opioid-Induced Respiratory Depression (OIRD) hours before a crisis. Bunkerhill Health's AI algorithms were cleared to quantify calcium on contrast-enhanced CT scans, now with a CMS reimbursement code. GuideAI Health also earned clearance for its software that detects peripheral vascular disease from CT scans, and PanopticAI secured clearance for contactless respiratory rate measurement via smartphone camera.
Why it matters
This flurry of clearances demonstrates a clear trend: AI is moving from a research concept to a regulated, integrated, and reimbursable component of clinical care. The shift from reactive alarms to proactive, predictive analytics (as in the Masimo case) is particularly significant, changing the paradigm of patient monitoring. For health-tech and robotics entrepreneurs, the establishment of a CMS billing code for AI analysis (Bunkerhill) is a critical milestone, creating a viable business model for AI-driven diagnostics.
Masimo describes its OIRD detection as a move from 'reactive alarms to proactive clinical intelligence.' Bunkerhill's new reimbursement pathway is seen as a key step to 'reduce barriers for health systems' to adopt AI. These clearances collectively show the FDA's increasing comfort with AI software as a medical device, especially for augmenting diagnostic imaging and physiological monitoring.
Researchers at MIT have developed 'Gleanmer,' a new system-on-a-chip that enables small, low-power robots to create detailed 3D maps of their surroundings in real-time. The chip consumes only 6 milliwatts of power, a drastic reduction compared to existing solutions that can require several watts. Gleanmer achieves this efficiency by co-designing the mapping algorithm and hardware. It uses a compact 'Gaussian occupancy' method to represent the environment, reducing memory needs, and incorporates specialized hardware accelerators to speed up processing.
Why it matters
This breakthrough solves a critical bottleneck for miniature robotics and lightweight AR/VR devices: providing sophisticated spatial awareness without prohibitive power consumption or memory requirements. By enabling tiny robots to navigate complex, unknown environments for extended periods, Gleanmer opens the door to new applications in areas like industrial pipe inspection, medical device simulation, and autonomous assembly of intricate products. For robotics developers, it represents a significant leap in deployable perception for power-constrained edge devices.
The MIT team emphasizes that previous attempts at low-power mapping were often too slow for real-time use, while real-time systems consumed too much power. Their solution, detailed on Tuesday, achieves both speed and efficiency by rethinking the problem from the algorithm up through the silicon. The use of Gaussian splatting for map representation, a technique recently popularized in computer graphics, proved to be a key innovation for compact and efficient spatial mapping.
Qualcomm is partnering with German firm Neura Robotics to integrate its new Snapdragon IQ10 processors into Neura's next generation of cognitive robots. The IQ10 is an AI-native System-on-Chip designed for high-performance, low-power edge computing, enabling robots to perform real-time sensing, reasoning, and action locally. The collaboration aims to bring more powerful on-device AI to both industrial and domestic robots, reducing their reliance on cloud connectivity.
Why it matters
This partnership signifies the growing importance of specialized silicon designed specifically for the demands of robotics. While NVIDIA has dominated the training and simulation space, Qualcomm is making a strong play for the inference-at-the-edge market. For robotics developers, this competition and the availability of powerful, energy-efficient, and integrated platforms like the IQ10 could simplify development, lower costs, and enable a new class of more autonomous and reliable robots that are not dependent on persistent network connections.
The move is seen as a direct challenge to Intel's and NVIDIA's positions in the edge AI market. The collaboration will leverage Neura's recent massive $1.4B funding round, which included Qualcomm as an investor. Neura's focus on 'cognitive robots' that can perceive and interact with their environment aligns with Qualcomm's 'agent-first' strategy of pushing AI processing to the device level.
Samsung Electronics announced on Tuesday the development of Universal Flash Storage (UFS) 5.0, the industry's fastest embedded flash memory. The new standard boasts read speeds of 10.8GB/s and write speeds of 9.5GB/s, roughly double the performance of UFS 4.0. The technology is specifically designed to handle the demands of on-device AI, enabling large AI models to operate directly on smartphones and other edge devices without requiring constant internet connectivity.
Why it matters
UFS 5.0 represents a critical hardware advancement for enabling powerful on-device AI. Faster data access directly translates to quicker inference times, reduced latency, and improved energy efficiency—all essential for real-time applications in robotics and autonomous systems. By removing the data-access bottleneck, this technology will allow for more complex AI models to run locally, enhancing the capabilities, privacy, and responsiveness of edge devices from phones to robots.
Samsung highlights that the improved performance and power efficiency will be key for the 'era of on-device AI.' This development complements the industry-wide push, led by companies like Apple and Qualcomm, to move more AI processing from the cloud to the edge. Mass production of UFS 5.0 is expected to begin in 2027.
Researchers have developed a reinforcement learning (RL) control strategy that allows swarms of microrobots to autonomously navigate and avoid obstacles in complex, unknown environments. According to a paper published Monday in Nature Machine Intelligence, the model uses a 'temporally extended attention' mechanism, allowing the swarm to consider historical sensor data to make better decisions. The system was trained in simulation using multi-level domain randomization to successfully bridge the 'sim-to-real' gap, enabling effective performance in real-world experiments involving tasks like cargo transport and dynamic target tracking.
Why it matters
This work addresses a fundamental challenge in microrobotics: moving from teleoperation in controlled lab settings to autonomous operation in unpredictable, real-world conditions. Enabling swarms to navigate intelligently without a pre-existing map is a critical step toward practical applications in fields like targeted drug delivery, minimally invasive surgery, and environmental monitoring. The successful sim-to-real transfer demonstrates a viable path for training complex swarm behaviors without exhaustive physical experimentation.
The research team highlights that their RL-based approach achieves or surpasses human-level performance in navigation tasks. A separate but related study in Nature Nanotechnology details the use of similar machine-intelligent algae-based microrobots ('algebots') for enhanced chemotherapy delivery in bladder cancer, showcasing the tangible medical applications of this technology.
Researchers at Virginia Tech have developed a method for creating durable, stretchable electronic circuits by ironing on liquid metal patches. Published on Tuesday, the technique uses a heat-activated adhesive to bond liquid metal encapsulated in a polymer shell directly onto fabrics. The resulting circuits are flexible, can withstand stretching, and remain functional after being integrated into textiles, opening up applications in health monitoring, environmental sensing, and robotics.
Why it matters
This innovation offers a simple, fast, and robust method for integrating complex electronics into soft and flexible materials, a key challenge in soft robotics. By enabling 'iron-on' circuits, the technology could dramatically lower the barrier to creating functional, multi-material soft robots and more sophisticated human-machine interfaces. It provides a practical way to embed sensing, actuation, and communication directly into the 'skin' of a soft robot.
The research team highlights the simplicity and accessibility of their method compared to complex fabrication processes like photolithography. The ability to create multi-layer circuits by simply stacking and ironing patches allows for more complex electronic designs. Another recent paper in Nature Communications describes using light to modulate surface textures on liquid crystal elastomers, further expanding the toolkit for creating dynamic, reconfigurable smart surfaces for soft systems.
A new perspective piece in Chemical & Engineering News argues that the primary bottleneck in advancing soft robotics is not engineering but fundamental polymer chemistry. The author contends that progress has been slow because the field relies on adapting existing polymers rather than designing new materials specifically for actuation. The article calls for polymer chemists to lead the development of next-generation materials to overcome limitations like high voltage requirements, slow response times, and poor durability in current soft actuators.
Why it matters
This analysis reframes the challenge of soft robotics, shifting the focus from mechanical design to materials science. For soft robotics to achieve its potential for creating life-like, compliant machines, a new class of purpose-built polymers is needed. This represents a significant opportunity for interdisciplinary collaboration and a call to action for materials scientists to design actuators from the molecular level up, which could unlock breakthroughs in the speed, strength, and efficiency of soft robotic systems.
The article suggests that current soft actuators are 'kludges' built from materials not intended for the task. It compares the situation to the development of OLEDs, which required decades of materials chemistry innovation. The author urges chemists to think of actuation as a primary design goal for new polymers, rather than an afterthought.
On Sunday, a DoorDash delivery robot named 'Dot' breached an active police perimeter during a SWAT operation in Chandler, Arizona. Despite crime-scene tape and verbal commands from officers, the autonomous robot continued on its delivery route, crossing the scene of a weapons investigation. A human technician was eventually called to intervene and manually remove the robot. The incident exposed a critical gap in the robot's ability to perceive and react to emergency situations.
Why it matters
This incident is a stark, real-world example of the 'edge cases' that plague autonomous systems. While robots can navigate predictable streets, they lack the contextual understanding to recognize and respond to dynamic, high-stakes human situations like a police emergency. This failure highlights the significant gap between current autonomous navigation and the general intelligence required for safe operation in public spaces. It will almost certainly trigger regulatory scrutiny and force developers to address how their systems identify and yield to first responders and emergency scenes.
Police on the scene expressed concern, noting the robot 'didn't know what to do' and created an unnecessary distraction. For its part, DoorDash stated it is investigating the incident. The event serves as a practical, and somewhat comical, illustration of the limitations of current physical AI, which excels at defined tasks but fails when confronted with the unscripted chaos of human activity.
Autonomous driving company WeRide, in partnership with automaker Geely Farizon and transport operator Kwoon Chung Bus Holdings, unveiled a new purpose-built, right-hand-drive (RHD) robotaxi on Monday. The vehicle was launched at the International Automotive & Supply Chain Expo in Hong Kong. The collaboration aims to mass-produce and deploy these robotaxis in RHD markets, beginning with Hong Kong, to address a significant gap in the global autonomous vehicle landscape.
Why it matters
The development of a purpose-built RHD robotaxi is a crucial step for the global expansion of autonomous mobility. It unlocks large, dense, and complex urban markets like Hong Kong, Singapore, the UK, and Australia, which have been largely unaddressed by the LHD-centric AV industry. This strategic move could accelerate the commercialization of robotaxi services worldwide by tailoring the technology to local traffic conditions and regulatory environments.
WeRide emphasized that this is not just a retrofitted vehicle but a ground-up design for mass production, ensuring consistency and reliability. Kwoon Chung Bus Holdings, a major Hong Kong transport operator, will provide valuable operational experience. The partnership combines WeRide's L4 autonomous driving technology, Geely's manufacturing scale, and Kwoon Chung's local market expertise, creating a vertically integrated model for deployment.
Policymakers at both the federal and state levels are making moves to create a clearer regulatory path for autonomous trucks. On Monday, a congressional subcommittee discussed the BUILD America 250 Act, a bill that proposes a national regulatory framework for autonomous vehicles. Separately, California has cleared the way for testing heavy-duty autonomous trucks on its public roads. These parallel efforts aim to provide the regulatory certainty needed for wider deployment.
Why it matters
The lack of a unified national framework has been a major impediment to scaling autonomous trucking in the U.S. These developments signal a serious effort to address the patchwork of state laws that complicates interstate operations. For the robotics and AV industry, a federal framework is crucial for long-term investment and deployment, while California's reversal of its previous stance opens up the country's largest state economy to testing and eventual operation.
Industry groups like the National Association of Manufacturers support the move, arguing it will strengthen supply chains. However, the proposal is facing pushback from groups like the Owner-Operator Independent Drivers Association. Testifying before congress, trucker Jonathan Wood argued for keeping human drivers in trucks, citing safety concerns with 'self-certified' autonomous technology and the risk of job displacement for hundreds of thousands of drivers.
Collaborative Robotics (Cobot) has launched Proxie Gen 2, its second-generation mobile robot designed for logistics, healthcare, and manufacturing. The new version features a higher payload capacity and introduces 'autotasking,' a capability that allows the robot to autonomously identify and execute tasks in its environment. The platform also offers an optional dual-arm manipulation system, integrating mobility with dexterity. The design is based on insights from over 13,000 hours of operation of the first-generation models.
Why it matters
Proxie Gen 2's 'autotasking' feature represents a step towards more truly autonomous systems that require less human intervention and pre-programming. By enabling the robot to identify its own work, it could significantly lower the complexity and cost of deployment, expanding the addressable market for mobile manipulators in dynamic environments where tasks are not always predictable. This moves beyond simple AMRs to more general-purpose, intelligent assistants.
The company, founded by former Google and Apple engineer Brad Templeton, is focused on creating robots that are useful 'out of the box' with minimal setup. The combination of autonomous task identification with mobile manipulation aims to solve the 'last meter' problem in logistics and other industries where robots need to not only move goods but also interact with them.
Humanoids Dominate Automate 2026 The Automate 2026 trade show has a dedicated 'Humanoid Robot Pavilion' for the first time, signaling a major industry shift from niche applications to a focus on general-purpose humanoids and the 'Physical AI' required to make them useful. Major players are showcasing commercial deployments, not just demos.
Safety Becomes a Product As robots, especially humanoids, move into production environments alongside people, safety is moving from a feature to a full-fledged product. NVIDIA's launch of 'Halos for Robotics' is a prime example, providing a full-stack, certifiable architecture to ensure safe human-robot interaction.
India's Affordable Humanoid Ecosystem Matures A cluster of startups in India, particularly around Bengaluru, are securing funding and launching humanoid robots priced for the domestic SME market. Companies like Astha, Huma, Indigo, Haven, and Agni are all targeting sub-$20,000 price points, creating a vibrant ecosystem for cost-effective industrial automation.
The Rise of the Microrobot Swarm Multiple research breakthroughs showcase the increasing sophistication of microrobotics, particularly in autonomous swarms. New reinforcement learning techniques are enabling swarms to navigate unknown environments, while biohybrid designs using algae are being developed for targeted drug delivery in cancer treatment.
Open-Source Infrastructure Gains Momentum Key new open-source projects are targeting foundational layers of robotics. Kyber, from the creator of VLC, is building a real-time control layer, while Rice University's OMPL 2.0 is democratizing advanced motion planning. This trend lowers the barrier to entry for developers and fosters community-driven innovation.
What to Expect
2026-06-27—Katalyst Space plans mission to boost NASA's Neil Gehrels Swift Observatory using its robotic spacecraft.
2026-07-01—China's new compulsory national standards for L3/L4 autonomous driving systems are set to be implemented.
2026-10-15—Opal Group's 'Transforming Healthcare with Emerging Technologies' forum begins.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
589
📖
Read in full
Every article opened, read, and evaluated
238
⭐
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