ACQUIRING TARGETS...
· DroidHunt news
Agility Robotics now expects to raise about USD 465M through its Churchill Capital Corp XI (CCXI) SPAC merger — down from roughly USD 620M in gross proceeds (USD 424M trust plus a USD 200M Foxconn-led PIPE) projected when the USD 2.5B pre-money deal was announced in June (Business Wire, Oct 13). The Robot Report reported the lowered expectation Oct 30, consistent with shareholder redemptions trimming the trust. CCXI shareholders vote on the merger Nov 2 at 9:00 am ET; if approved, the combined company trades as AGLT and the close is expected later in November (DroidHunt, Oct 14).
Why it matters: Lower trust cash leaves a thinner runway after redemptions — a standard SPAC outcome, but one that raises the bar on Digit 5 converting the USD 300M+ order book into deployed, revenue-generating robots through 2026–27.
Sources: therobotreport.com money.tmx.com
Tags: agility-robotics ipo us reported
Hongqi Chain, China's first listed convenience-store chain, terminated its robot unmanned-vending R&D project in July after 16 months, saying the system fell short of mass-commercialization standards; 2025 R&D spend on it was about RMB 2.87M (~USD 400k). QbitAI ROBO field reporting published Nov 1 finds retail humanoids remain largely proof-of-concept: 100+ robot retail outlets now operate across 30+ Chinese cities, but a store setup costs nearly RMB 1M (~USD 140k), robots mostly just fetch items in 20-plus seconds, human minders are still required for restocking and charging, and no vendor ships 100+ units into a single retail scene. Insiders estimate 2–5 years to human-comparable clerk work.
Why it matters: A listed retailer publicly calling time on robot retailing is rare, hard evidence of where the capability line sits. It tempers the retail-deployment narrative behind several humanoid business plans and supports a factory-first sequencing of deployments.
Sources: qbitai.com
Tags: hongqi-chain market-data china reported
First-order policy learning — differentiating through the simulator itself — has rarely worked for contact-rich humanoid control because gradients explode. ∇Mimic (arXiv 2610.41654, posted Oct 30) removes both failure modes with median-relative per-sample gradient clipping and multistep regression onto gradient-derived action targets. Its open-source NablaEngine differentiable GPU simulator for the Unitree G1 computes full MuJoCo contact dynamics with analytic gradients 11.5× faster than the MuJoCo Warp adjoint. One policy on one GPU reaches the state-of-the-art whole-body tracking score (0.82) in 20 minutes versus 1.2 hours for the leading world-model method, and 0.874 in 1.8 hours, across all 40 LAFAN1 clips. SHAC, the standard first-order baseline, stops on exploding gradients within minutes.
Why it matters: Moves contact-rich humanoid skill training from GPU-days toward minutes on a single GPU and releases the engine as open infrastructure. Cheaper, faster whole-body reinforcement learning directly lowers the data-generation cost embedded in every humanoid roadmap.
Sources: arxiv.org
Tags: nabla-mimic research other reported
A Zhejiang University–led team with NUS and industry collaborators posted Movens-Zero (arXiv 2610.41877, Oct 30): a humanoid motion foundation model built on Movens-5M — 15M human-motion sequences mined from raw internet video and refined into 5M physically executable, execution-verified humanoid-motion sequences spanning 10,000+ hours, aligned with text, RGB, speech and music. A single 911M-parameter action expert, trained with task-mixed flow matching across seven generation tasks, generates motion from any of the four modalities, with multimodal real-time chunking for closed-loop control. Scaling execution-verified data from 1% to 100% and parameters from 64M to 911M improved results monotonically across all modalities.
Why it matters: Internet video is the largest untapped motion-data source, and execution verification is the filter that makes it usable. Monotonic scaling across four input modalities suggests a training-data path that does not depend on teleoperation fleets.
Sources: arxiv.org
Tags: movens-zero research china reported
Home Every digest Robot catalog Drones on DronePriceHunt Privacy & affiliate disclosure RSS