Robotics developments — 23–30 May 2026

Matternet announced approximately $33 million in an oversubscribed private placement alongside a completed reverse merger with Los Altos Ventures, which was renamed Matternet. The transaction makes the drone-delivery developer subject to U.S. public-company reporting requirements; the financing is a public-market placement rather than a venture round.

New investors included Ed Eisler of EE Holdings and Mark Tompkins of Montrose Capital Partners. Proceeds will support its next-generation aircraft platform and commercial expansion in food, retail and healthcare. Matternet combines delivery drones with ground stations and fleet software, building on its FAA aircraft approvals and existing healthcare and restaurant delivery operations.

Linkerbot disclosed a completed Series B+ at a $3 billion valuation, without publishing the amount raised. Reuters reported participation from Zhongguancun Science Park Fund, Bank of China Asset Management and Fosun Capital. The company separately sought a $6 billion valuation for its next financing, without specifying its timing or whether it would be private or an IPO.

Linkerbot supplies dexterous robotic hands for industrial manipulation and research, including customers mounting them on existing robot arms. CEO Alex Zhou said production was close to 5,000 hands a month, with a target of 10,000. Its LinkerSkillNet platform converts recorded human manipulation skills into reusable robot-hand behaviours, connecting the component business with training data and control software.

Ondas reported more than $30 million in new orders during May, taking second-quarter orders to date above $110 million. The bookings cover air defence, counter-drone systems, reconnaissance, unmanned ground vehicles and other autonomous defence and security equipment.

The company is integrating these businesses through shared engineering, manufacturing, customer access and operational support so customers can combine multiple systems within a mission. The figures describe orders supporting future backlog conversion, rather than revenue already recognised. Ondas separately reported $50.1 million in first-quarter revenue and a $457 million pro forma backlog.

Tycho.AI secured a $2 million, 18-month AFRL Tactical Funding Increase contract for TRIDENT, a programme developing maritime and overland autonomy in environments with unreliable or unavailable GPS. The award advances its Voyager hardware and software stack through flight tests, data collection and integration with existing operational systems.

Voyager combines visual perception, navigation and control on compact onboard computing. The work includes rotary- and fixed-wing aircraft trials, autonomous waypoint traversal, coordination between platforms and resilient perception, with integration into FANTOM middleware. TRIDENT builds on a $1 million AFWERX SBIR Phase II grant awarded in 2024; the new contract funds development and validation rather than a production fleet purchase.

Volatus Aerospace advanced to the next evaluation phase of the U.S. Drone Dominance Program, having submitted a one-way-attack FPV platform for the long-range-strike mission area. Its entry draws on allied industrial collaboration and the company’s manufacturing, autonomy and operating capabilities.

The next stage evaluates operational performance and readiness for scalable production. Selection gives Volatus a route into a U.S. defence procurement programme, but it is not a signed purchase order. The wider programme’s funding and intended drone volumes should therefore remain separate from revenue or orders awarded to Volatus.

Figure signed a commercial agreement with Catalyst Brands to introduce humanoids into its distribution and logistics network, beginning at the Reno, Nevada Distribution Logistics Center. The initial work targets physically demanding supply-chain tasks at a retailer whose portfolio includes JCPenney, Aéropostale and Brooks Brothers.

The agreement creates a route from warehouse demonstrations into a named customer’s operations. Brookfield is an investor in Figure and a backer of Catalyst, providing a commercial connection between the companies. Figure did not disclose a robot count or deployment timetable in the announcement, so the agreement should not be read as a delivered fleet figure.

NASA awarded Lunar Outpost a $220 million performance-based, firm-fixed-price task order for the first phase of Lunar Terrain Vehicle Services. The contract supports development of its Pegasus rover and a planned 2028 lunar delivery through NASA’s Commercial Lunar Payload Services programme.

Pegasus evolves the company’s Eagle design and is intended to carry astronauts under manual control while also supporting autonomous and teleoperated operation between crewed missions. NASA described a vehicle capable of more than nine miles per hour and up to a year of lunar operation. Work over the following 18 months covers design maturation, crew evaluations and flight-unit qualification, ahead of the planned mission.

Singapore’s Civil Aviation Authority awarded a Thales-led consortium with Deeeplabs a contract for a national unmanned-aircraft traffic-management system. Based on TopSky–AstraUTM, the platform will provide a single portal for approval and tracking of drone operations across Singapore.

The system combines operator and aircraft registration, digital flight planning and authorisation, and real-time monitoring. Thales supplies the traffic-management platform while Singapore-based Deeeplabs supports deployment and operational services, bringing regulatory workflows and flight oversight into the same infrastructure.

Terra Drone subsidiary Unifly was selected as the technology partner for the operational traffic-management component of Malaysia’s ClearPath UASTMS national drone platform. The U2 system will monitor and coordinate drone flights in real time, supporting oversight as unmanned operations expand.

Datasonic, a NexG subsidiary awarded the national programme in 2025, is developing U1, the registration layer. Unifly contributes flight-operation management through U2, connecting that administrative infrastructure with the tools used to manage actual aircraft movements. The Malaysian selection is separate from Thales’ Singapore contract.

Eden Robotics recorded Eden I as designed, manufactured and operational on May 23, saying development took one and a half months. The milestone followed its Theta OS v0.2.1 software going live on the first demonstration robot on April 3.

The company is developing robot hardware and software for autonomous industrial services, with Eden I presented as an early semi-humanoid platform. The update documents completion of an operational prototype; it does not identify a customer installation or establish commercial production volumes.

Astribot announced T1, a semi-humanoid service-robot platform starting at CNY89,900. The company’s launch post describes cable-driven manipulation, human-like dexterity and customisable AI computing, positioning the robot for vision-language-action models and service applications.

The platform gives developers hardware on which to build manipulation behaviours and adapt computing to their chosen models. The launch establishes the announced starting price and product positioning, without supplying task-success measurements or confirming a customer deployment.

LimX Dynamics introduced Luna, a full-size interactive humanoid intended for shopping malls, theme parks, exhibitions and brand events. The 160 cm robot has 27 body degrees of freedom, excluding end effectors, and uses the company’s System 0 whole-body motion model to coordinate movement.

The platform supports learning motions from video, a no-code task editor and synchronised control of multiple robots. Those tools let operators create and update performances and interaction programmes for commercial spaces. The announcement introduces a product and its content tools; it does not establish a fleet of customer deployments.

Twolabs introduced Tobi, a semi-humanoid robot under development for nursing homes and senior-living communities. Its intended applications include feeding, dressing, medication reminders, fetching objects, walking support, wayfinding and companionship. The company reported that its prototype design was complete while the AI system remained in active development.

Twolabs is building the hardware alongside vision-language-action models and a social-interaction layer. Its aim is to combine physical assistance with conversation and support for residents, reducing repetitive work for care staff. These are planned capabilities and applications, rather than demonstrated care deployments.

EngineAI opened its approximately 12,000-square-metre Honghualing manufacturing base in Shenzhen and announced the first batch of T800 humanoids leaving the line for delivery. The company described the facility as establishing 10,000-unit scalable delivery capability, rather than reporting 10,000 robots already shipped.

The plant combines incoming-material inspection, component assembly and testing, final integration, end-of-line tests, shipment and after-sales maintenance. EngineAI states that the layout can produce a robot every 15 minutes. The new base is intended to make manufacturing and quality control repeatable as T800 moves from prototypes into larger customer batches.

Kraken Robotics reported first-quarter revenue of C$21.7 million, up 35% year on year, and adjusted EBITDA of C$3 million. Product revenue increased 50%, led by demand for SeaPower subsea batteries and synthetic-aperture sonar. The company said announced product orders for 2026 had reached approximately C$97 million.

Its batteries and sensing systems support underwater vehicles and seabed inspection, while KATFISH combines towed sonar with launch and recovery equipment. Kraken maintained standalone 2026 revenue guidance of C$165–175 million and adjusted EBITDA guidance of C$40–50 million. Those forecasts excluded Covelya, whose acquisition remained pending and was expected to close near the end of the second quarter.

Arbe Robotics reported first-quarter revenue of $0.5 million, compared with $0.04 million a year earlier, and a $9.4 million net loss. It also reported follow-on Phoenix radar orders from robotaxi customers and initial chipset shipments to Hirain for an L4 automated-driving project with a Chinese automaker.

The company supplies imaging-radar technology for detecting objects around vehicles, including systems intended for 360-degree sensing. Hirain is developing a second, lower-channel-count platform alongside its existing 48-by-48 imaging radar. Arbe retained 2026 revenue guidance of $4–6 million and said future automotive design wins would depend on customer adoption cycles, with no timetable given for additional awards.

Genesis released version 1.0.0 of its open-source physics simulation framework, adding non-convex multi-contact collision detection and a mesh-repair pipeline. These changes help simulations represent contact between irregular shapes, which matters when testing robot manipulation and movement around real objects.

The release also accelerates the rigid constraint solver, tactile point-cloud sensing and the material-point-method solver. The team reported rigid-solver improvements of up to 35% in contact-rich scenes with about 64 degrees of freedom. That result describes the tested simulation workload; the release also fixes contact drift and unwanted rotation to improve the behaviour developers use for robot testing and training.

AGIBOT introduced BFM-2, a generative motion-control model that plans transitions from a robot’s current physical state to a target configuration. Its two-stage Motion-Between architecture uses the robot’s dynamics, contact conditions and higher-level instructions to generate movement, including recovery from disturbed or unusual postures.

The company demonstrated transitions from lying, leaning and other states, presenting the model as a way to maintain continuous movement when higher-level commands change. BFM-2 addresses the control layer beneath task-planning models. AGIBOT also outlined a future BFM-3 incorporating additional sensory inputs; that roadmap was separate from the released model’s demonstrated capabilities.

Researchers introduced COBALT, a cloud-based teleoperation system for collecting robot demonstrations using smartphones and other commonly available controls. The architecture shares compute across users and streams video and control data, allowing people in different locations to operate simulated or physical robots.

The study reported sub-100 ms end-to-end latency with up to eight concurrent users per GPU and separately demonstrated 256 simulated clients across eight GPUs. A five-day pilot collected more than 7,500 smartphone demonstrations, exceeding 50 hours, from nine countries. User training and automatic quality filtering supported the collection, and the researchers evaluated the resulting data by training imitation-learning models.

X Square Robot open-sourced Wall-OSS-0.5, a vision-language-action model, alongside weights, code, training recipes and optimisation tools. Its training combines proprietary robot demonstrations, open data from multiple robot embodiments and 90 million multimodal examples to improve manipulation before task-specific fine-tuning.

The company evaluated the pretrained model on 17 physical tasks without task-specific tuning. Reported task-progress scores included 100 for block sorting, 96 for fruit sorting, 86 for ring stacking and 82 for rope tightening; these are progress scores rather than successful-run percentages. A separate fine-tuned evaluation across 15 tasks averaged 60.5 progress, providing a second setting for developers assessing the released model.

microAGI launched Shift’s free apartment-cleaning service in New York City, with customers agreeing to record the work for robot-training data. Vetted operators wear recording devices while completing normal household tasks; the value of the first-person footage funds the cleaning service.

Founder Bercan Kilic said personal information is anonymised before processing. The programme gathers demonstrations of physical work in varied homes, adding real-world task examples to the company’s data collection. Handyman work, repairs and errands were described as future services, while the launch centred on cleaning.

Pronto confirmed a small opt-in pilot recording cleaning and other household jobs, according to original reporting by Moneycontrol. The home-services company said the experiment involved a very small set of users and that recordings were deleted within 48 hours.

The pilot explores recording physical workflows in customer homes as demand grows for human-activity data used in AI and robotics. Privacy experts questioned how consent, retention and reuse for model training would be governed. The reported activity remained a limited experiment, with no wider recording rollout established.

Wayve launched Wayve Labs, a research unit led by chief scientist Jamie Shotton, to develop embodied intelligence for autonomous vehicles and other physical systems. Its programme covers world models, reinforcement learning, spatial understanding, simulation and large-scale policy learning.

The lab will use Wayve’s driving data, deployed fleets and simulation infrastructure to study how models can plan actions and learn from their consequences. Research priorities include uncertainty, physical interactions, rare safety-critical situations and transfer of learning between different robotic systems. The announcement establishes a research programme extending beyond driving without introducing a separate commercial robot product.

Fraunhofer IPA introduced a modular benchmark for industrial humanoids, covering basic and complex capabilities, cleanroom suitability, functional safety, cybersecurity and energy efficiency. Manufacturers and users can commission individual modules or a full evaluation to assess whether a robot meets the requirements of a particular application.

The first comprehensive assessment used a Unitree G1 EDU-4 with Dex3-1 hands, delivered in May 2025 and running firmware 1.04. Tests found good self-stabilisation but collision forces exceeding 500 newtons and a Bluetooth vulnerability that has since been fixed. Battery runtime reached two hours 49 minutes standing and one hour 49 minutes in mixed standing and walking. The results describe that tested configuration and provide application-specific evidence for deployment decisions.

ARGUS Interception agreed to integrate Ouster digital lidar into its A1-Falke net-based drone interceptor. The collaboration adds short-range three-dimensional sensing to improve target tracking during the final approach and capture, particularly around critical infrastructure and public spaces where destructive interception can create additional hazards.

A1-Falke uses an interception drone with two net launchers and a sensor suite combining radar, lidar and depth cameras. It is designed to capture an uncooperative drone and bring it down intact for recovery and inspection. The companies also plan to explore Ouster’s Rev8 lidar for future capabilities, including native-colour sensing and longer-range detection.

XTEND was selected for the Phase II qualifier of the U.S. Department of Defense’s Drone Dominance Program. The summer event at Camp Grayling, Michigan, will evaluate systems in complex and contested scenarios. Selection provides access to the next assessment stage and does not represent a procurement award to XTEND.

The company plans to demonstrate XOS, its operating system for human-guided autonomous robots and drones. XOS combines onboard autonomy with operator control for missions in hazardous environments. The wider programme aims to support procurement of more than 200,000 drones by 2027, a programme target that is separate from any allocation to individual participating companies.