Green by Sberbank vs Cruzr S2 by UBTECH Robotics

Green is a full bipedal humanoid at 180 cm and 75 kg, with 41 degrees of freedom including 12 finger joints and 4 wrist joints. Cruzr S2 puts its humanoid upper body on wheels instead of legs. Green's hands, with force sensing in each fingertip, are built for detailed manipulation.

Cruzr S2 is more than twice as fast on flat floors, at up to 2 m/s against Green's 0.9 m/s walking speed. A wheeled base is simple and stable in shops and lobbies. Green walks like a person and navigates unfamiliar indoor spaces on its own, steering around obstacles.

Green has eight RGB cameras and two LiDARs for vision, inertial sensors for balance and force-torque sensors in its fingertips. These give it the perception for both navigation and handling small objects such as test tubes or retail items.

Green's AI combines Sber's GigaChat, which lets it hold conversations and follow voice commands, with the Green-VLA vision-language-action model (about 1.3 billion parameters, open weights) for manipulation. It can spot and correct its own mistakes, and AI agents can give it a personality for different roles.

Sber has said Green uses a 48 V battery but has not published its runtime or charging time. A wheeled robot like Cruzr S2 generally spends less energy moving around than a walking humanoid, which matters for long days on a shop floor.

Green is aimed at collecting items and stocking shelves, assembling and packing products, lab work with test tubes and bio samples, and room service and cleaning, with catering also named by Sber. Cruzr S2 is a commercial service robot suited to flat indoor spaces where quick wheeled movement matters more than walking.

Green relies on inertial and force sensors for balance, avoids obstacles and can detect its own errors; Sber has not published safety certifications. A wheeled base keeps Cruzr S2 low and stable. Ask both makers for certification details before deploying near the public.

Green's software is built on Sber's GigaChat and the open-weight Green-VLA model, and Sber's Green Challenge contest lets outside teams train control models in a simulator of the robot. Sber has not announced a public SDK.