MarkTechPost
Kyutai, with Inria and École des Ponts, released REKI, a 2.4B-parameter vision-language-action (VLA) model for robot manipulation. It is pretrained only on public data and keeps a cheap memory of its recent past. On three out-of-distribution benchmarks, only models about twice its size beat it, and all weights are public.
Why it matters: Robot policies often act on the current frame alone, so lightweight memory is a practical gap. A small open model that is competitive with larger ones lowers the barrier for robotics labs that cannot train on proprietary data. It adds to the push to make embodied models cheap enough to run on-device.
Ars Technicabig story
A company called Scaleout is deploying decentralized AI-driven learning to military bases and drones, using small on-device models that let drones identify and attack battlefield targets without a human in the loop at the moment of engagement.
Why it matters: Autonomous target selection and engagement is one of the most contested lines in military AI policy and international law, and running the models directly on the drone makes it viable even without reliable connectivity. This will intensify debate over autonomous weapons just as AI researchers themselves report growing concern about extreme risks from advanced AI.
Ars Technica
Agility Robotics unveiled a humanoid robot designed to stop or crouch when it detects a nearby human coworker, rather than requiring a safety cage. The goal is to let the robot work directly alongside people on a factory or warehouse floor.
Why it matters: Removing the cage requirement is the main practical barrier to deploying humanoid robots in existing warehouses and factories without costly retrofits, so proximity-safety behavior like this matters more for real deployment speed than raw dexterity specs.
MarkTechPost
Reward AI released OM-1, a general-purpose manipulation policy trained solely on human demonstrations captured via a 7-degree-of-freedom wearable glove, with no teleoperation or on-robot data used. The policy runs on industrial arms and humanoids at human speed and can learn a new task from under 30 minutes of data. No weights, code, or API have been released publicly yet.
Why it matters: Most robot learning today still leans on expensive teleoperation rigs or robot-specific data collection; if human-demo-only training scales, it could sharply cut the cost of teaching robots new tasks. It fits a broader push toward general-purpose robot policies that generalize across embodiments rather than being trained per-robot, alongside other recent robotics gains.
TechCrunch Startups
Maven Robotics came out of stealth with a $100 million Series A, saying it already has active robot deployments and is positioning itself to compete for robotics deployment contracts with established players.
Why it matters: Unlike many robotics startups still at the prototype stage, Maven claims live deployments already in place, suggesting the deployment-and-integration layer of robotics is becoming its own competitive market — similar to how systems integrators emerged around earlier waves of industrial automation.
The Decoder
On a new robotics benchmark called StationeryBench, GPT-6 Astra completed 7 out of 100 dual-arm robot tasks, while competitor MolmoAct2 completed zero. A researcher described the jump as a "step change" in spatial reasoning.
Why it matters: Even a single-digit success rate is notable here because general-purpose language models have historically struggled with embodied, physical-world tasks. If this generalizes, it suggests frontier LLMs are starting to close the gap with specialized robotics models, a prerequisite for more capable real-world agents.
NVIDIA
Skild AI's S1 robot foundation model, built with Nvidia's Physical AI tools, can learn previously unseen, long-horizon tasks from a single video demonstration rather than extensive reprogramming. It targets manufacturing and warehouse settings where layouts and tasks change frequently.
Why it matters: One-shot task learning from video could meaningfully cut the integration cost that has kept general-purpose robots out of dynamic environments, where reprogramming for each new task has been the main barrier to adoption. It's part of a broader push toward foundation models for robotics, mirroring how LLMs generalized across language tasks.
TechCrunch
Maven Robotics came out of stealth with a $100 million Series A funding round. The startup already has active robot deployments in place.
Why it matters: Landing paying deployments before a public launch signals real commercial traction rather than pure hype, a rarer feat in the crowded robotics-startup field. It adds another well-funded challenger competing for deployment contracts that larger robotics players currently hold.
MarkTechPost
Axis Robotics released AXIS, a browser-based data collection system that moves robot demonstration gathering out of physical labs onto GPU-backed backends. The dataset includes 207 manipulation tasks and 50,129 verified trajectories on a Franka arm. Pretraining on it lifted the pi-0.5 robot policy's LIBERO-Plus score from 83.9 to 88.8, versus 57.5 for a volume-matched control dataset.
Why it matters: Robot learning has lagged language models partly because collecting real demonstration data requires physical hardware and lab access; a browser-based pipeline could let far more contributors add trajectories, similar to how crowdsourcing scaled text and image datasets. The gap over the volume-matched control suggests data quality and diversity, not just quantity, is driving the improvement.
Tom's Hardware
At the 2026 World Humanoid Robot Games in Beijing, the Tiangong Ultra robot ran 100 meters in 9.39 seconds, and the Honor Lightning followed at 9.47 seconds — both beating Usain Bolt's 9.58-second world record. The Tiangong Ultra reportedly reached 23.8 mph before crashing into padded barriers at the finish.
Why it matters: Raw sprint speed isn't itself commercially useful, but it's a visible benchmark of how fast humanoid robot locomotion and actuator control are progressing, particularly from Chinese robotics programs pushing hard on physical performance alongside AI-driven data centers and chips. Expect this kind of engineered stunt to keep showing up as a proxy metric in the broader US-China robotics competition.