All articles
FLUX-mimic: a video-action model cuts robot training to 30 minutes
mimic robotics and Black Forest Labs unveiled FLUX-mimic, a video-action model that cuts task training from 30+ hours to about 30 minutes. It already runs on an Audi line handling soft-body manipulation.
Brain waves as robot training data — Encord's experiment
Encord and Germany's Zander Labs are testing operators' brain waves as data for training physical AI. EEG and muscle sensors aim to capture error, intent and surprise — signals invisible on video.
Stationary OP1 humanoid handles 30% of orders at Highline's hub
Highline Commerce deployed stationary OP1 humanoids from Ultra Robotics that now handle about 30% of daily order fulfillment at its Brooklyn hub. The robots run on Physical Intelligence's π0.6 foundation model.
Figure builds its 1,000th Figure 03 humanoid as output scales
Figure's BotQ plant has built its 1,000th Figure 03 humanoid. The US maker hit a pace of one robot per hour and targets 50,000 units a year — still far behind its Chinese rivals.
NEURA Robotics opens NEURA Gym at RWTH Aachen for physical AI
NEURA Robotics and RWTH Aachen open NEURA Gym — a physical AI training facility for robots. It is one of 10 gyms meant to close the real-world data gap.
Generative Bionics' Gene.01: a humanoid with full-body touch
Italian startup Generative Bionics, spun out of IIT, unveiled Gene.01 — a humanoid covered in full-body tactile sensors. It targets industry, with a Fincantieri shipyard as the first proving ground.
Holiday Robotics raises $105M for FRIDAY wheeled humanoid
Seoul startup Holiday Robotics raised $105M in Series A for its FRIDAY wheeled humanoid. The robot bets on dexterous hands — 40 of its 64 degrees of freedom sit in the hands — targeting repetitive factory work.
NASA runs Google's Gemma 3 in orbit to analyze satellite imagery
NASA has run a large language model in orbit for the first time — Google's Gemma 3 on the YAM-9 satellite analyzes Earth imagery in real time and downlinks text instead of raw images.
Runway launches an AI model router for generative media
Runway launched Media Router, a tool that automatically picks the best model for image, video or audio generation by priority: quality, speed or cost. It answers the fragmentation of the generative media market.
Claude voice mode gains Opus and Sonnet models
Anthropic opened Claude's voice mode to Opus and Sonnet models — previously it ran only on Haiku. The update rolls out in beta across all platforms, with integrations for Gmail, Slack and Notion.
AMD takes on Nvidia with Helios rack-scale AI system
AMD unveiled Helios, a rack-scale system built on Instinct MI450 GPUs, and a deal with Anthropic to deploy up to 2 GW. Lisa Su says the rack beats Nvidia's Vera Rubin.
Nvidia sends Jetson to the Moon — autonomous rovers cut the delay
Nvidia is sending its Jetson chips to the Moon. Lunar Outpost's autonomous rovers will test them on missions launching before the end of 2026 on a Falcon 9. The chip processes sensor data locally so the rover can decide without waiting on Earth.