← Zen Chronicles

Robotics Chronicle

Pilot issue · Embodied AI Wednesday, September 23, 2026 14 Reads · 8 Schoolhouse · 3 Hype-sieved
Boundary: robots and physical systems — not pure LLM policy (that stays on The AI Chronicle). Substance first; demo reels last. Manual pilot labels · taxonomy robotics-v1.

solid|high_yield + analysis|explainer|research_summary — start here.

RESEARCH SUMMARYmodels_for_robots

HuRo: Robotizing Human Videos for Scalable VLA Pretraining

Topicmodels_for_robotsGenreresearch_summaryLearninghigh_yieldStanceneutralSubstancesubstance

Paper: pipeline robotizes heterogeneous human videos into aligned obs+actions; HuRo ~630K episodes / 142M frames. Real-world ALLEX tasks: completion 51.5%→80.3% with scale; OOD 34.9%→72.2%; visual overlay and action retargeting both matter.

Read or skip: SCHOOLHOUSE. READ: best recent evidence that robotizing human video (obs+actions) scales real-robot completion and OOD robustness.

Original →

RESEARCH SUMMARYsafety

ShieldVLA: Feasibility-Aware Safety Alignment for Vision-Language-Action Models

TopicsafetyGenreresearch_summaryLearninghigh_yieldStancecautiousSubstancesubstance

Paper: HJ-reachability safety critic gates VLA fine-tuning; rubric-based VLM safety scores; reports ~57% lower cumulative safety cost and +0.13 task success vs SafeVLA across five nav/manipulation benchmarks (sim).

Read or skip: SCHOOLHOUSE. READ abstract+intro if you care how VLA fine-tuning can gate on HJ reachability instead of soft Lagrangian penalties. Sim-only for now.

Original →

RESEARCH SUMMARYmodels_for_robots

Zero-WAM: In-Context World-Action Modeling from Human Videos for Open-Ended Task Generalization

Topicmodels_for_robotsGenreresearch_summaryLearninghigh_yieldStanceneutralSubstancesubstance

Paper: treats human video as in-context task specification for manipulation; HumanGen 74.2K human-robot ICL pairs; reports 47.0% avg success on seven unseen RoboTwin 2.0 tasks (+29.5 pp vs strongest video-action baseline) plus real-world generalization notes.

Read or skip: SCHOOLHOUSE. READ if you want the in-context human-video prompt angle (Zero-WAM) alongside HuRo’s robotization story.

Original →

ANALYSISgeopolitics_supply

Who wins and who loses after US bans foreign robots?

Topicgeopolitics_supplyGenreanalysisLearningsolidStancecautiousSubstancesubstance

FCC adds foreign-produced advanced robotic devices to Covered List (Jul 28). Covers mobile humanoids/quadrupeds/AMRs with sensors+connectivity; exemptions for drones, medical, fixed industrial arms, prior-authorized models. Analysts split on whether ban helps US makers or starves labs of cheap platforms.

Read or skip: SCHOOLHOUSE. READ: what the FCC Covered List actually covers/exempts, and why some analysts say it may starve US labs of cheap platforms.

Original →

ANALYSISindustrial_labor

Humanoid Robots Are Now Picking Products in U.S. Warehouses. Here’s How That Compares to China.

Topicindustrial_laborGenreanalysisLearningsolidStancecautiousSubstancesubstance

GXO CEO Patrick Kelleher (FreightWaves Today, Aug 12) says 45+ humanoid pilots and live cosmetics/apparel picking; operating cost $10–15/hr needs to fall below $10 (~2 years). Compares US Digit unit costs vs China shipments and notes July 2026 FCC foreign-robot Covered List move.

Read or skip: SCHOOLHOUSE. READ: rare operator-side ROI talk ($10–15/hr → <$10) plus US vs China cost/shipment contrast. Treat China % figures as cited secondary, not audited.

Original →

ANALYSISproducts_demos

World humanoid robot games show runners breaking records, bursting into flames

Topicproducts_demosGenreanalysisLearningsolidStancecautiousSubstancesubstance

Viral sprint records coexist with crashes, sparks and fires. IEEE-affiliated researcher Dipam Patel argues many robots were engineered for one task (including inability to brake), while laundry/hotel/firefighting scenarios and teleoperation penalties reveal where autonomy still lags.

Read or skip: SCHOOLHOUSE. READ: clearest teardown of Games spectacle (no braking, task-specific bots, teleop penalties) vs duller chore/hotel/fire tests that matter more.

Original →

ANALYSISmodels_for_robots

Figure’s Helix 2.5 Takes on Chores in 30 Unseen Homes

Topicmodels_for_robotsGenreanalysisLearningsolidStancecautiousSubstancesubstance

Independent read of Figure’s Helix 2.5 numbers: 237/420 successes (~56%); clarifies zero-shot applies to homes/objects not untrained chores; notes Sunday Robotics Tony Zhao’s reliability critique and non-comparable ACT-2 fold metrics.

Read or skip: SCHOOLHOUSE. READ after r05: defines zero-shot correctly, tables the 420-trial split, and surfaces Zhao’s reliability objection without inventing a bake-off.

Original →

ANALYSISindustrial_labor

CJ Logistics Puts Humanoid Robots on a Live Packing Line

Topicindustrial_laborGenreanalysisLearningsolidStancecautiousSubstancesubstance

Cautious trade-desk writeup of CJ Logistics’ Sep 3 announcement: two dual-arm humanoids insert cushioning on a live Olive Young pack line near Seoul. Partners Robotis / Aidin / RLWRLD; no cost or throughput numbers; ‘first in Korea’ treated as company framing.

Read or skip: SCHOOLHOUSE. READ for a skeptical desk take on a live but narrow pack-line toehold — and explicit call-out that ‘first in Korea’ is company framing.

Original →

Useful wires and lab/field detail — read cautiously, not as proof of mass deployment.

NEWS DISPATCHproducts_demos

China robot makers seek to turn humanoid hype into useful work

Topicproducts_demosGenrenews_dispatchLearningmodestStanceneutralSubstancemixed

Chinese robot makers showed humanoids sorting parcels, packing phones and doing chores at the World Robot Conference in Beijing, seeking a path from spectacle to commercial use. Unitree shares soared nearly sixfold on Shanghai debut; Nvidia’s Madison Huang visited; deployments cited at China Post and Lens Technology.

Read or skip: READ IF tracking China deployment anecdotes and Unitree IPO heat. SKIP IF you already know WRC is still mostly booth demos — learning payoff is modest.

Original →

NEWS DISPATCHproducts_demos

Robots can outrun humans, but can they plug in a cable?

Topicproducts_demosGenrenews_dispatchLearningsolidStancecautiousSubstancesubstance

At Beijing’s World Humanoid Robot Games, robots beat human sprint records while organizers also staged warehouse, cable-connection and other work scenarios. More than 40% of events require full autonomy per Huawei; industry voices stress reliability over running and jumping.

Read or skip: READ for the cable/warehouse framing against sprint spectacle. Not Schoolhouse genre (dispatch), but the best single wire on ‘demo ≠ work’.

Original →

NEWS DISPATCHhardware

After outrunning Bolt, China's robot champion races towards real-world work

TopichardwareGenrenews_dispatchLearningsolidStancecautiousSubstancesubstance

Exclusive with X-Humanoid lead engineer Jack Guo after TianGong Ultra’s 8.64s 100m. Focus shifts from race programs trained in simulation to reliability, braking, and early factory box-moving trials; commercial deployment counts undisclosed.

Read or skip: READ for workshop-level detail on Ultra’s sim race programs and Guo’s reliability pivot. SKIP pure sprint highlight reels.

Original →

Vendor launches and spectacle previews. Skim for facts; discount the theater.

Helix 2.5: Zero-Shot 30-Home Generalization

Figure AI · 2026-09-17 · promo_launch · mixed

HYPE SIEVE (vendor promo). Skim numbers only; pair with r06 before trusting ‘zero-shot homes’ headlines. 56% full-task ≠ household product.

Original →

Agility Unveils Digit 5 Humanoid Robot Built for Cooperatively Safe Work at Scale

Agility Robotics · 2026-09-15 · promo_launch · mixed

HYPE SIEVE (vendor launch). Note real field hours claim and 2027 availability; discount order-book theater and CE-mark forward language until independent coverage.

Original →

2026 World Humanoid Robot Games to Be Held August 22–26

YouTube · 2026-08-01 · promo_launch · hype

HYPE SIEVE (spectacle preview). Watch only as visual accompanying r02/r03 — not a learning piece.

Original →

Robotics Chronicle pilot for Zen Ridge · 14 curated items from ~last 30 days where possible · classifications manual (no OpenRouter spend) · taxonomy at chronicles/robotics/taxonomy-locked.json · links go to originals · no invented quotes · text-only v1 (no xAI audio this issue).