1080pGemini Robotics: coordinating whole-body movement
An official whole-body demonstration shows a robot carrying a container while moving, linking instruction understanding with execution.
Official model demonstrationRead case
LEARNING & SIMULATION
Embodied intelligence & physical AI
Study how robots learn from demonstrations, language and interaction, alongside the data and code.
Google DeepMind · official video frameFollow the path from demonstration collection to policy execution on specific robots and tasks. Cases distinguish autonomy, teleoperation and simulated results.
PUBLIC CASES
1080p
GR00T N1: bimanual humanoid manipulation
Studying pick-and-place, handovers and bimanual coordination with hardware demonstrations, simulation and task-specific adaptation.
Model and hardware manipulation experimentsRead case
π₀: from language instructions to laundry handling
A cross-embodiment vision-language-action model studies folding laundry, clearing tables and assembling boxes.
Hardware manipulation researchRead case
720pMobile ALOHA: learning with two arms and a mobile base
Teleoperated demonstrations support learned mobile tasks such as opening cabinets, rinsing a pan and cooking.
Hardware imitation-learning experimentsRead case
1080pHOVER: switching whole-body control modes within one policy
A humanoid policy tracks hand, head and root commands while coordinating upper and lower body motion.
Hardware control experimentsRead case
VideoAtlas: autonomous handling of engine-cover parts
Moving parts between supplier containers and a sequencing dolly from a list of bin locations.
Manufacturer hardware demonstrationRead case
1080pEELS: a snake robot for passages through ice
A segmented robot combines rotating screw propulsion with environmental mapping in ground tests on ice, snow, sand and glacier passages.
Ground prototypes and glacier testsRead case
540pTAP: torque, angle and pressure for tactile feedback
Combined tactile measurements provide contact-state feedback for placement, stacking and cutting experiments without vision.
Manipulation experimentsRead case
JOURNAL
Field-record notePerception & localisation
The next move beneath the ice
EELS brings locomotion, mapping and risk assessment into a single exploration task.
Methods discussionHumanoid robotics
What can we learn from humanoid robotics?
Start with control, perception and data, then examine access to a specific space.
SELECTED SOURCES
Robot policies, world models and embodied reasoning serve different roles. This index includes open resources and official model records, with access and disclosure stated for each. Task capability must be assessed with the robot and experimental conditions.
Browse Learning & adaptation ↗Working on a related study? Submit a project with its objectives, progress and resource access.
Visit community projects ↗