ACQUIRING TARGETS...
· DroidHunt news
Boston Dynamics has finalized the supplier lineup for its Atlas humanoid, led by Hyundai Mobis, and has asked selected suppliers to prepare for purchase orders covering 10,000 units, KED Global reported Oct 28. The lineup is entirely Korean — a mobilization of the Hyundai Motor Group parts base behind commercial production of Atlas, extending the Hyundai deployment storyline (DroidHunt, Sept 23). Earlier reporting targeted Atlas production deployment around 2028; no timeline or pricing accompanied the PO preparation. Neither Boston Dynamics nor Hyundai Mobis has publicly confirmed the report.
Why it matters: This is the first concrete unit-scale procurement signal for Atlas and shows Hyundai converting its Atlas commitment from pilot programs into supply-chain reality. A 10,000-unit frame also sets a benchmark against which earlier 25,000-unit deployment pledges can be measured.
Sources: kedglobal.com
AgiBot released Genie Master, an embodied-Agent framework the company says is the first to run Physical RSI (recursive self-improvement) in a real business setting rather than simulation or lab toys. Deployed on AgiBot Lingxi G2 units in a working convenience store, the system lets robots diagnose their own failures, adjust strategy, and bank the experience without engineers in the loop: a wrist camera compensates for a head-camera blind spot on a failed restock, for example. Two loops do the work — Harness RSI updates prompts, memory, and rules while keeping model weights fixed, and Policy RSI retrains the underlying action policy on filtered real-world data. It is the fourth release in AgiBot's "AI WAVE" series that began Oct 22. A 24-hour livestream with six G2 robots staffing a store is scheduled for Nov 2. All performance claims are company-provided.
Why it matters: Per-site engineering and data collection are the biggest recurring costs in commercializing humanoids. If deployment-time self-improvement works at scale, it compresses the cycle from "customer requests a task" to "robot can do it" — the core economics of fleet deployment.
Sources: qbitai.com
Tesla VP of Vehicle Engineering Lars Moravy and Cybercab lead Eric Earley said in a Munro Live interview that Optimus humanoids will plug Robotaxis into Superchargers as a stopgap until wireless inductive charging is ready for the fleet. Humans currently perform the task by hand in Austin. Tesla holds FCC approval for wireless Cybercab charging and is planning an Austin robotaxi hub with 80 wireless chargers; the executives added that Optimus might eventually service vehicles. It is the first stated role for Optimus inside the Robotaxi workflow, following the start of Optimus production at Fremont (DroidHunt, Oct 22).
Why it matters: This is the first named, structured operational job for Optimus outside Tesla's own factories — a constrained, high-repetition environment that suits current-generation robots and could function as a commercial on-ramp ahead of general availability.
Sources: notateslaapp.com
Tags: tesla deployment us reported
Google DeepMind announced Genie 4 on Oct 27, a natively multimodal world model that generates real-time interactive simulations with synchronized audio, ambient sound, and dialogue, built on Gemini Omni anchored by Gemini-family reasoning. DeepMind explicitly positions training environments for robots and agents as a core use case alongside interactive media. Rollout starts as a research preview for trusted testers, then reaches Project Genie for Google AI Ultra subscribers and early-access partners of the Gemini API and AI Studio.
Why it matters: Cheap, infinitely variable simulated environments are a direct input to robot policy training and evaluation. Genie 4 competes with NVIDIA Cosmos-class world models for the simulation layer of the embodied-AI stack — a layer where training cost and coverage increasingly differentiate robot programs.
Sources: deepmind.google
RoboParty has published the full-stack open-source release of its RP1 humanoid (DroidHunt, Oct 4): URDF/MJCF models, mechanical and electrical design documentation, embedded control interfaces running 500 Hz control and 50 Hz policy updates, and the hhtools human-motion-retargeting toolchain, plus a live GitHub repository and a beta co-creation program. RP1 stands 135 cm with 24–30 active degrees of freedom depending on configuration, built on RoboParty's ROMOMO joint modules (25–160 N·m peak torque, 48 V, CAN-FD). CAD files, BOMs, and PCB manufacturing data remain partially internal, and licensing terms for commercial reuse are not yet stated.
Why it matters: Most "open" humanoids stop at SDKs and APIs. Full-stack openness down to joint design gives research labs reproducible hardware and could make RP1 a common testbed for embodied models — provided the repository is maintained and the license permits real reuse.
Sources: qbitai.com
← Newer: Oct 29, 2026 Older: Oct 27, 2026 →
Home Every digest Robot catalog Drones on DronePriceHunt Privacy & affiliate disclosure RSS