Where AI meets robotics: humanoid robots and labs racing to ship them, embodied foundation models, warehouse and industrial deployments, and the investor and policy backdrop.
A longitudinal exploratory field study evaluated whether a software robot can enhance cognitive functions in older adults. The paper frames the work against rising global dementia prevalence and situates the trial within the expansion of digital cognitive training as a scalable nonpharmacological intervention.
Agility Robotics announced Wednesday that it will merge with Churchill Capital Corp XI in a transaction valuing the company at $2.5 billion, according to reporting by the Associated Press and TechCrunch. The deal is expected to provide more than $620 million of proceeds, with roughly $200 million coming from a group of new and existing institutional investors and the remainder from Churchill Capital and public investors, TechCrunch and GeekWire report. Agility's flagship humanoid, Digit, is in commercial use at a small number of customer sites and the company says it has secured more than $300 million in multi-year orders for the next-generation Digit v5, per TechCrunch. Early customers and backers named in reporting include Amazon, Toyota, Schaeffler, Mercado Libre, Nvidia, SoftBank, and Foxconn.
An Associated Press feature profiles a wave of startups and researchers pivoting from language models toward so-called world models - AI systems designed to learn the statistical structure of physical space, time, and interaction. Computer scientist Louis Castricato left his PhD at Brown University to found Overworld, which in January 2026 released Waypoint-1, a real-time interactive video diffusion model trained on 10,000 hours of game footage, backed by a $4.5M pre-seed from Kindred Ventures. AI pioneer Fei-Fei Li describes "world model" as "one of the most important and most overloaded terms in AI today," per the AP. The piece contrasts language-only models with systems designed to model light, motion, and physical interaction, and profiles entrepreneurs experimenting with robotics, simulation, and embodied deployment.
Greek robotics startup Acumino has raised $11.7m in seed funding, per Sifted. The company was selected by Google DeepMind for its first European robotics accelerator cohort (announced June 9, 2026) alongside 14 other early-stage European robotics companies. Acumino builds hardware-agnostic physical AI for dexterous tasks in industrial automation and plans to use the seed capital to scale commercial customer deployments, according to Sifted. The DeepMind accelerator is a three-month, equity-free program providing access to Gemini robotics models, up to $350,000 in cloud credits, and technical mentorship from DeepMind engineers, culminating in a Demo Day in London in September 2026.
Hyundai Motor Group has announced plans to buy SoftBank Group's remaining 9.65% stake in Boston Dynamics for $325 million, a purchase that would make the robotics firm a wholly owned subsidiary (Reuters, June 19; UPI, June 23). Hyundai originally acquired an 80% controlling stake in 2021 in a deal valued at $1.1 billion; subsequent capital increases reduced SoftBank's holding to about 10% (UPI). Multiple Hyundai affiliates - Hyundai Motor, Kia, Hyundai Mobis and Hyundai Glovis - each hold Boston Dynamics shares and must complete their own board approvals; a Hyundai official cited around July 20, 2026 as the deadline for SoftBank's put option (UPI; Asia Today). KED Global links the buyout to a potential Boston Dynamics IPO and to corporate governance restructuring within Hyundai Group (KED). SoftBank's exit lets it redeploy capital toward AI infrastructure bets (StartupFortune).
Defence Secretary Dan Jarvis confirmed at the Royal United Services Institute Land Warfare Conference on 23 June that the forthcoming Defence Investment Plan will invest in uncrewed ground vehicles, coverage and a transcript of his speech indicate; UK Defence Journal quotes Jarvis saying, "The DIP will make real those ambitions, and that includes investing in the uncrewed ground vehicles the Army requires to build the next generation of land forces." Reporting by Forces News and Forces.net records that General Sir Roly Walker, the Chief of the General Staff, argued at the same conference that future crewed deployments should be accompanied by uncrewed ground vehicles and outlined the Army's recce-strike integration of autonomy, AI and precision fires. Industry context: This publicly stated procurement intent moves uncrewed ground systems toward formal programme status rather than niche experimentation, according to public coverage.
According to RealSense's press release and product page, the company unveiled the D585 Pro depth camera at Automate 2026. The device combines a custom Gen 5 SoC with on-camera AI acceleration, a depth engine, image signal processor, DSP and a quad-core Arm processor, per RealSense. Key specs listed by RealSense include a 120deg x 100deg field of view, depth output up to 1280 x 960 at full resolution, full-resolution operation at 60 FPS and depth frame rates up to 90 FPS. RealSense's product materials and reporting from The Robot Report list a minimum close-range capability around 10 cm at full resolution and an ideal operating range beyond 10 m, with a maximum exceeding 20 m. The Robot Report also states the D585 Pro is expected to begin shipping in Q1 2027. Editorial analysis: the combination of wide FOV, extended range and on-device inference accelerates deployment of perception stacks in AMRs, humanoids and inspection robots.
Researchers at Zhejiang University developed a vision-based control system that enables a quadcopter to navigate narrow openings using only onboard vision and sensors, according to Hackster.io. The team trained a neural network policy with reinforcement learning in simulation and then transferred it to a real vehicle, per Hackster.io. Instead of a conventional pipeline of state estimation, gap detection, trajectory planning, and control, the system maps camera images and other onboard sensor data directly into low-level flight commands, Hackster.io reports. The custom test platform measured 38 cm wide and 10 cm tall and carried a monocular camera, a PX4 flight controller, and an NVIDIA Jetson Orin NX, according to Hackster.io. To accelerate exploration, the researchers used trajectories produced by a model-based planner to guide the learning process, per Hackster.io.
Jeonbuk National University (JBNU) is positioning physical AI at the core of its global competitiveness strategy, per a Korea Times report. The university has been selected to lead South Korea's national Physical AI Proof-of-Concept program, a project expected to total nearly 1 trillion won (~$730M) over five years from 2026. An 846-sq-meter manufacturing technology test lab is already operating, showcasing synchronized industrial robots, with a 181,000-sq-meter physical AI campus set to begin construction in 2027. Hyundai Motor Group, Naver and other corporate research centers are expected to participate in the broader program.
Hackster reports that maker Becky Stern modified an open-source Backyard Brains SpikerBot to detect tabletop edges and avoid falling. According to Hackster, Stern combined a mirror and a spiking neural network based on the Izhikevich-style model to drive the robot's responses. The article lists the bot's onboard components, including a ESP32-S3 microcontroller, OV2640 camera module, VL53L4CD laser distance sensor, microphone, accelerometer, RGB LEDs, and continuous-rotation servos, and notes Stern documented the full build on Hackster.
Sonar+D's 2026 exhibition moves into Barcelona's historic Llotja de Mar, combining immersive art, robotics and AI across 38 works, per the festival's programme page. Highlights reported by Domus and Sonar include Astral Twin by Brooklyn studio Volvox Labs - two robotic arms performing in dialogue with a giant LED wall (created with OFFF) - and Qs Ventures 2147: A Voice from the Future, an abandoned phone booth hosting an AI persona representing Earth as a living superorganism. Belgian studio Superbe's from0 installation transforms recorded sounds into melody via 12 pendulums. Designboom (active media partner) also notes Google DeepMind's Lyria 3 features in the sonic programme. Domus captured visitor-facing moments and a quoted exhibitor describing a video project called "Bardo" as "a kind of in-between state."
The Kurzweil Library republished in June 2026 a landmark 2015 UC Berkeley study on BRETT (Berkeley Robot for the Elimination of Tedious Tasks), a PR2 robot that learned dexterous motor tasks through deep reinforcement learning without task-specific programming. The original research, led by Pieter Abbeel and Trevor Darrell with postdoc Sergey Levine and PhD student Chelsea Finn, demonstrated BRETT completing tasks such as placing a clothes hanger, stacking rings, assembling a toy airplane, screwing a bottle cap, and inserting a peg, using a unified algorithm for all tasks. Abbeel stated: "What we're reporting on here is a new approach to empowering a robot to learn. The key is that when a robot is faced with something new, we won't have to reprogram it." Presented at ICRA 2015, this work was a milestone in applying deep RL to real-world robotic manipulation and influenced subsequent sim-to-real and multi-task transfer research.
In a Q1 2026 earnings recap published by 24/7 Wall St. via Yahoo Finance, Tesla CEO Elon Musk warned investors that early Optimus and Robotaxi ramps will start 'quite slow.' Tesla reported $22.39 billion in Q1 2026 revenue (+16% year over year), beating non-GAAP EPS estimates at $0.41 vs. $0.3481 expected. CFO Vaibhav Taneja confirmed over $25 billion in 2026 capital expenditures targeting robotics and autonomy, while operating expenses rose 37% on AI R&D. Musk said of Optimus: 'it is literally impossible to predict, except that I think it will be quite slow for us as we iron out the 10,000-plus unique items.' Prediction markets (Polymarket) assign 15% odds to an Optimus consumer release by year-end 2026 and 3% to a California Robotaxi launch by June 30.
MicroAGI, a German startup, offers free home cleaning in New York City through its Shift app, recording first-person video of cleaners at work and using the footage to train and license AI/robot datasets, according to The Verge, Business Insider, Ars Technica, Android Authority, and Entrepreneur. Business Insider published a first-person account from a reporter who hosted Shift cleaners and a visiting chef in her apartment; the reporter said the workers wore head-mounted cameras and that she hid personal items before the visit. MicroAGI's website says it anonymizes 'names, faces, or other personal information.' Entrepreneur reports the broader operator program spans '10,000+' contributors across '15+' countries and that operators earned over $5 million in Q1 2026, a figure Entrepreneur attributes to the company.
VLC Media Player lead developer Jean-Baptiste Kempf raised $5 million for Kyber, a Paris-based startup building an open-source SDK for controlling remote machines in real time, per TechCrunch and The Next Web. The round was led by Lightspeed Venture Partners, with OVNI Capital and Kima Ventures also participating. Kyber synchronizes video, audio, sensor data, and control inputs at minimal latency - the company demonstrated 8 milliseconds glass-to-glass at the Mile High Video conference in February 2025. Kempf told TechCrunch the platform targets 'all the use cases where the person who's operating is not in the same place as the compute, which is not in the same place as the action.' The startup is already in commercial deployment with customers in defence, telco, robotics, and AI. Kempf frames scale as the central problem: the largest remote driving fleets today manage around 2,000 to 3,000 vehicles, but anticipated million-vehicle robot fleets require a different infrastructure layer entirely. Kyber's core SDK is open source; the company sells enterprise licenses and deploys forward-deployed engineers for custom integrations.
Researchers published Sashimi-Bot in npj Robotics, a three-armed robotic system that prepares sashimi from a raw salmon loin. Three manipulators divide roles: one arm holds and slices with a chef's knife fitted with a GelSight tactile sensor, a second stabilizes the fish, and a third picks up slices with chopsticks. Manipulation policies were trained via deep reinforcement learning in simulation and transferred to the real robot without additional real-world tuning, per the paper. The GelSight sensor, trained on 12,397 readings from 157 cutting motions, achieved 95% accuracy in detecting blade-to-board contact (99% precision, 67% recall), per ZME Science coverage. In full tests, the system cut 34 salmon slices and successfully placed 26 of 28 that landed on the board; six slices that stuck to the blade were all recovered.
AFP-JIJI reporting in The Japan Times from the Robotics Summit in Boston notes that humanoid robots can mix cocktails, run marathons and fold laundry, but many remain teleoperated or restricted to scripted chores. The report cites prototypes and products including Elon Musk's Optimus, Figure AI's Figure 03, China's AgiBot and Matrix Robotics, and 1X's Neo, the latter described as being steered by a person (AFP-JIJI in The Japan Times). Engineer and vendor quotes include Chris Matthieu of RealSense, who said, "Most of the humanoids you see are being teleoperated, or they've got very specific paths and chores that they do," and William Okazaki of Renesas, who said AI has "extremely accelerated" growth (AFP-JIJI). Editorial analysis: Industry practitioners should view current humanoids as advancing in specific skills, not as general-purpose assistants.
The Korea Herald, Korea Times, and Seoul Economic Daily report that LG CNS and Doosan Corp. signed a memorandum of understanding on June 18, 2026 at LG Science Park in Seoul to pursue cooperation across agentic AI, robotics, data centers, and cloud. Korea Herald reports LG CNS President Hyun Shin-gyoon and Doosan Corp. President Yoo Seung-woo attended the signing. Reporting covers two LG CNS AI platforms: enterprise agentic AI platform AgenticWorks and robot learning platform PhysicalWorks, both to be integrated with Doosan's industrial capabilities. The firms will establish a joint cooperation task force within one month. Korea Herald notes the companies have been partnered since 2025 on hydrogen fuel cell-based logistics, with this MOU extending into AI and robot transformation.
Aether AI announced the closing of a $20 million seed financing round, led by MPCi with participation from Inno Angel Fund, SWC Global, Unity Ventures, and others, per a press release distributed via GlobeNewswire. Founded by Prof. Biwei Huang of UC San Diego, the company is building Causal World Models for Physical AI and robotics, positioning causal reasoning rather than statistical correlation as the foundation for reliable AI systems. The round will fund R&D, team expansion, and initial commercial deployments in robotics. Prof. Huang presented the company's Four-Layer Causal Brain Architecture at CVPR 2026 in Denver, where internal benchmarks showed 20-30% improvements in data efficiency on manipulation tasks vs open-source baselines (Aether AI; vendor-reported). Research lineage includes Causal-Learn and Causal-Copilot, widely adopted open-source causal AI tools created by Huang, and the company cites advisory relationships with prominent researchers including Judea Pearl and Bernhard Scholkopf.
South Korea's Industry Minister Kim Junggwan visited Prague on June 17-18 to chair the second Dukovany Project Implementation Review Committee and the third Korea-Czech Supply Chain and Energy Dialogue. The Dukovany nuclear power plant project - awarded to Korea Hydro & Nuclear Power (KHNP) and described as one of the largest energy infrastructure initiatives in Central Europe - is progressing on schedule, per the Korea Times. Ministers reviewed joint R&D in robotics, batteries, and semiconductors, and Minister Kim visited the Czech Technical University robotics testbed. The EU recently closed its state-aid probe into Czech government support for the project, clearing a key regulatory hurdle. Both sides confirmed intent to extend the nuclear partnership to the Temelin plant.
Alibaba has launched the Qwen Robot Suite, a set of three foundation models for robotics developed by its Tongyi Lab, according to reporting by the South China Morning Post and Alibaba's product blog (Qwen). The suite includes Qwen-RobotNav (vision-language navigation), Qwen-RobotWorld (a video world model for prediction and simulation) and Qwen-RobotManip (a generalist vision-language-action model), per company materials reported by SCMP and PYMNTS. SCMP reports Qwen-RobotManip was trained on more than 38,100 hours of open-source data and topped the RoboChallenge generalist track with a process score of 59.83 and 45% task success rate, according to Alibaba. The models are in pilot testing with selected Alibaba Cloud enterprise customers, SCMP and Yahoo News report.
BetaKit reports that Autonomique, a Montreal- and Menlo Park-based startup, has moved beyond pilot testing and is deploying its physical AI software at a F&P Manufacturing production line in Tottenham, Ontario. According to BetaKit, Autonomique's system directs semi-humanoid robots to assemble chassis and suspension parts that enter vehicles in fewer than four hours. The company integrates its software into third-party hardware rather than building robots, BetaKit writes. BetaKit also reports Autonomique closed the first tranche of an undisclosed seed round backed by investors including Innovobot, Garage Capital, Inovia Capital, and the co-founders of Clearpath Robotics and Otto Motors. In an interview with BetaKit, co-founder and CEO Vikrant Tomar said, "You will see robots doing backflips or dancing...but it's useless when it comes to actual productivity."
NVIDIA GEAR Lab, in collaboration with Carnegie Mellon University and UC Berkeley, has released ENPIRE, an agentic robotics framework that lets frontier coding agents autonomously learn dexterous manipulation tasks on real robot hardware - including GPU insertion into a motherboard - with no human intervention, per the official NVIDIA research page and trade coverage. Linxi 'Jim' Fan, NVIDIA's Director of AI & Distinguished Scientist and GEAR Lab co-lead, described the system as enabling AutoResearch in the physical world for the first time. ENPIRE comprises four modules: Environment (EN), Policy Improvement (PI), Rollout (R), and Evolution (E). Coding agents - including Codex (GPT-5.5), Claude Code, and Kimi Code variants - were tested in fleet sizes of 1, 4, and 8 robots, achieving a 99% success rate on challenging tasks such as GPU insertion, pin organization, and zip-tie cutting. The NVIDIA research page indicates an open-source release is planned.
AFP and multiple outlets report that Indian workers are recording first-person household tasks to train AI systems that guide robots. Workers such as Nagireddy Sriramyachandra film chores wearing head-mounted smartphones or cameras and are paid about 250 rupees per hour (approximately $2.6), according to AFP, Al Jazeera and NDTV. Data firm Objectways, cited in reporting by The Economic Times and NDTV, collects these "egocentric" videos for clients and uses platforms including Amazon SageMaker. Objectways CEO Ravi Shankar is quoted listing requested tasks such as "folding clothes, coffee making... sandwich making," per The Economic Times. Reporting also cites a Morgan Stanley projection that there could be over one billion humanoid robots in use by 2050. Digital labour expert Aditi Surie is quoted saying data collection services are likely to increase, per Al Jazeera.
Showing 24 articles
