Valkyrie
A robust, rugged, entirely electric humanoid robot designed to operate in degraded or damaged human-engineered environments.
Primary use case: General-purpose humanoid robot for performing versatile tasks that emulate human capabilities without a human inside (e.g., maintenance of human-designed facilities such as offshore oil rigs)
Specifications
Physical
Locomotion
bipedal
Manipulation
Hand Type
dexterous
Power & Autonomy
Autonomy Level
unknown
Commercial
Availability
unknown
Primary Use Case
General-purpose humanoid robot for performing versatile tasks that emulate human capabilities without a human inside (e.g., maintenance of human-designed facilities such as offshore oil rigs)
Related news
Robot ambulances, robot protests, and China says the US is now copying them
Sanctuary AI built a humanoid. Its best industrial worker so far is a pair of hands.
nasa.gov
Manual entity-ingest supporting URL (2026-10-24)
Gemini 4 Argon Writes Physics Simulation Code From Human Video, Teaches Apollo Humanoid Cornhole
Google DeepMind says Gemini 4 Argon can turn ordinary human video into MuJoCo simulation code and a robot policy, enabling Apptronik's Apollo humanoid to land seven consecutive cornhole shots without bespoke robotics fine-tuning. The article also reports a tabletop task-completion improvement from 56% to 75%, an 87% score on Google-designed ERQA, and cautions that the demonstrations have not established broad generalization to complex industrial manipulation.
Google Cloud, NASA Test AI-Powered Robot for Deep-Space Missions
Google Cloud and NASA are testing an AI-powered robotic system designed to handle cargo, equipment organization and medical-support tasks during future deep-space missions, where Earth communications may be delayed by up to 44 minutes on a Mars mission. In an early proof-of-concept test at Johnson Space Center's iMETRO facility, an open-source Pi0 vision-language-action model with 4 billion parameters located and moved a medical bag with an 85% success rate after being fine-tuned in about 2.5 hours on Google Cloud G4 instances, and the resulting model can run locally on compact edge hardware.