Embodied AI

The FCC proposed on Tuesday to restrict imports of humanoid and animal-shaped robots on national security grounds. China's Commerce Ministry condemned the move on Thursday as discriminatory and threatened countermeasures. The US argues these robots could become data-collection tools as they enter homes and workplaces, and sees advanced robotics as a strategic technology. Chinese manufacturers dominate the global humanoid robot market, with startups already selling units below $5,000 and producing most key components domestically. In 2024, over 2 million robots were operating in Chinese factories, with 300,000 new installations—more than the rest of the world combined. The post does not disclose the FCC proposal's timeline or voting schedule.

Gemini Robotics 2 is a vision-language-action model that outputs full-body joint commands, not just single-arm control. In a demo, Apptronik's Apollo 2 robot picks a baseball glove off a shelf, coordinating fingers to feet. Google says it used 1,040 preference samples for alignment to make motions more natural. It's lab-only for now; the post doesn't disclose a commercial timeline or pricing.

A White House interagency body ruled on July 27 that all foreign-produced advanced robots pose an unacceptable risk to U.S. national security and should be placed on the FCC's Covered List. The determination covers quadrupeds, bipeds, wheeled, and tracked devices, citing their high-fidelity sensors and always-networked nature as vectors for data exfiltration and remote hijacking. It flags three domains—critical infrastructure patrol, manufacturing, and military UGVs—while offering foreign producers a transitional path via conditional approvals as they onshore production. The document does not specify an effective date or name affected manufacturers.

Epoch and METR's MirrorCode benchmark shows Claude Opus 4.7 reimplemented a 2–17 week human coding task in 14 hours for $251, though it still struggles with projects like ruff. Anthropic had Opus 4.7 autonomously finish robot fetch tasks in 9 minutes 35 seconds, 20x faster than last year's human-assisted record. Robot startup Sunday confirmed the same pattern: scale pretraining, then fine-tune on small high-quality data, hitting 99.1% on laundry folding.

A weekend rumor claimed Anthropic was buying robotics startup Physical Intelligence. The CEO denied it quickly. Physical Intelligence, co-founded by Lachy Groom, has raised over $1B and was reportedly in talks for another $1B round at an $11B valuation. Its π0.5 model is widely used in robotics research. The article confirms the two sides did hold acquisition talks, but doesn't disclose terms or why they broke down. The rumor spread fast partly because both Anthropic and OpenAI have been on acquisition sprees this year.

Xiaomi Robotics open-sourced XR-1, a ready-to-use robot foundation model. It pre-trains on 100K hours of embodiment-free manipulation videos across 1,700+ scenarios, then post-trains on 7,200 hours of real-robot data for embodiment and instruction alignment. Pre-training shows clean scaling laws—lower action error with more data and larger models—and those gains transfer directly to real-robot success rates with no sign of saturation yet. After post-training, XR-1 picks up new tasks like phone packing and printer refilling from under 10 hours of demos on average, hitting 75% overall success (nearly 2× π 0.5); with under 40 hours it reaches 85%. It also achieves SOTA on four sim benchmarks. Code, weights, and paper are public.

Huang spent July 15–16 in Tokyo locking in deals that turn Japan from a chip-material supplier into a full-stack physical-AI partner. The centerpiece is Noetra: 44 domestic firms led by SoftBank, Sony, NEC, and Honda, backed by ¥1 trillion ($6.2B) over five years, building homegrown AI for robots, vehicles, and factory floors. Nvidia also launched the Cosmos robotics alliance with Fanuc, Yaskawa, and others to train robots on simulated data. On the materials side, JSR, Shin-Etsu, and Tokyo Electron secured next-gen AI chip supply deals. Worth flagging: these are framework agreements, not shipped products. But the structure—Nvidia wiring itself into Japan's entire industrial stack—matters more than any single contract.

Agility Robotics leased a 60,000 sq ft facility in Fremont, California to train its Digit humanoid robots, just up the road from Tesla's planned Optimus factory. CEO Peggy Johnson says Digit is already generating revenue at Amazon, GXO, Schaeffler, and Toyota's Canadian plant, with $300M in contract orders claimed. The post doesn't disclose how many Digits are deployed; outside observers estimate dozens. I'd take the $300M figure with a grain of salt until we see actual shipment volumes.

NVIDIA announced two new Thor-based modules: T3000 (865 FP4 teraflops, 32GB memory, 273GB/s bandwidth) at roughly half the size and power of T5000, and T2000 (400 FP4 teraflops, 16GB) for broader edge AI. New Jetson agent skills automate memory optimization—some customers saved up to 15GB and moved to lower-memory SKUs. Cosmos 3 Edge, a 4B-parameter world model, runs on-device on Thor for real-time vision and robot policies. The post does not disclose pricing or ship dates for T3000/T2000.

Per LatePost and supply chain sources, Tesla issued parts procurement guidance for Optimus: ramp to 1,000 units/week by September and 2,000–2,500/week by year-end, implying ~100k units/year capacity. At a late-June exec meeting, Musk approved the final Optimus Gen 3 design and said he'd fire the entire procurement team if the year-end target isn't met. The Fremont factory's former Model S/X line is now the robot line; concrete orders for hundreds of units in August are already placed. Musk himself tempered expectations, saying initial production will be 'extremely slow' because everything is new and the robot involves ~10,000 unique parts.

Mistral released Robostral Navigate, an 8B model that lets robots navigate indoors and outdoors using only a single RGB camera. It skips lidar and HD maps, outputting velocity and steering angle directly from visual input. The post doesn't disclose latency, frame rate, hardware requirements, training data size, or benchmark details. I'd hold off on excitement until we see third-party tests.

General Intuition just closed a $320M round at a $2.3B valuation, with Coatue, Eric Schmidt, and researchers from MIT and Google DeepMind joining. CEO Pim de Witte argues on the Equity podcast that LLMs like ChatGPT and Claude lack spatial-temporal understanding—gaming data fills that gap. Eight minutes of real-world data was enough to get a robot navigating an office cold. The company turned down an acquisition offer reportedly from OpenAI and built Nerve, a marketplace connecting gamers to data labeling and teleoperations work to get ahead of AI-driven job displacement.

Forterra disclosed it has deployed more than 100 self-driving ATVs in Ukrainian conflict zones over the past nine months, the largest known combat use of autonomous ground vehicles by a US defense tech firm. The vehicles handle logistics and recon. A company exec said no defense tech is proven until it hits real combat. Ukraine turned to ground autonomy because aerial drones have made soldiers extremely vulnerable. The post does not disclose weapon configurations or loss figures.

Yann LeCun argued at VivaTech that LLMs like ChatGPT can't handle real-world complexity and aren't a path to human-level intelligence. His new venture AMI Labs, founded after leaving Meta in 2025, is building JEPA—an architecture that learns abstract representations instead of memorizing statistical patterns. The company raised over $1B in seed funding from Nvidia and Jeff Bezos' family fund. Oxford's Ingmar Posner is pursuing a similar direction with world models that reason about causality. The post does not disclose JEPA's performance benchmarks or a product timeline.

NVIDIA's ENPIRE lets physical robots self-improve through trial and error like coding agents, hitting 99% on tasks like GPU insertion and zip-tie cutting. The catch: auto-evaluation and auto-reset still break on harder tasks. Tencent open-sourced ARGUS, an always-on tracing system for 10k+ GPU training clusters, already battle-tested for six months. A separate law paper points out that top minds badly misjudged nuclear fission and the internet—today's AI hot takes will likely age just as poorly.

General Intuition raised $320M to train AI on millions of hours of gameplay footage. Founder Pim de Witte argues that keystrokes, mouse movements, and decision sequences teach models physical reasoning better than text. They plan to sell the resulting models to robotics firms and game developers. The post does not disclose valuation or investor names.

AWS released the Strands Robots SDK under Apache 2.0, wrapping the LeRobot stack into a unified agent. It defaults to MuJoCo simulation with no hardware needed; switch to mode="real" for physical robots. Recorded demos are saved as LeRobotDataset and can be pushed to Hugging Face Hub. Policies like GR00T or LerobotLocal run inference, then broadcast commands to multiple robots over Zenoh mesh. Simulation and hardware code are identical except for one keyword argument. Examples run in a notebook with Python 3.12+ on Linux/macOS, no GPU required.

Qwen team released Qwen-RobotManip, a foundation model for robotic manipulation. The key insight: alignment, not just larger pretraining, is what makes scale pay off. Demos show cross-embodiment generalization across real robots—stacking bowls, folding clothes, making burgers, arranging flowers—with Qwen-Omni issuing open-ended voice commands on the fly, no predefined task list. The post does not disclose model size, training data scale, or latency figures; only demo videos and a paper link are provided.

Prometheus raised $12B at a $41B valuation. The startup targets automating heavy engineering and drug design in the physical world. The post only discloses the round size and valuation—no details on tech approach, team, or how the money will be spent.

Huawei Cloud announced CloudRobo, an end-to-end embodied AI platform that covers data, model training, deployment, and integration on a petabyte-scale trusted data base. At INSPIRE2026, partners showed dual evaluation for data and models, fast assembly of active force-control models, robot cloud onboarding in hours, and model deployment in minutes. The post doesn't disclose pricing or regional availability—I'd hold off on the 'world's first' claim until we see production deployments.