Jumper 22-DOF Open-Source Crab Robot

Description

Coming soon — not yet available for purchase

Jumper is a pre-release robot. KingKong Technology has not announced a price or a retail launch date, and no retail channel is open yet. We are tracking the project and will open orders once the retail version ships. Everything below is taken from the manufacturer and the official open-source project.

Jumper — “跳跳” in Chinese — is a 22-degree-of-freedom crab robot from KingKong Technology, a Shenzhen company founded in 2017 whose magnetic encoders sit inside most of China’s humanoid robots. It walks on six legs, jumps more than 400 mm, grips up to a kilogram, and shows its mood on two round displays. The interesting part is not the shape: it is that the toolkit used to teach it every one of those movements was published as open source.

Jumper 22-DOF open-source crab robot in red on a dark background with green light streaks

What it is

A six-legged indoor robot measuring 400 × 400 × 200 mm and weighing 1.8 kg. Twenty-two tactile servos drive it, each carrying dual magnetic encoders so the controller knows both the motor position and the joint position — the feedback that makes a 400 mm jump possible at all. The brain is a Rockchip RK3576 with an 8-core CPU, a 6 TOPS NPU, 4 GB of RAM and 64 GB of storage, fed by a dToF depth sensor, a 162-degree 5 MP camera, a 6-axis IMU, four microphones and two speakers.

Demo modes you can train for include a tripod gait, walking, posture, gestures, beat-synced dancing, jumping and two-claw grasping. The trade route is indoor only, with about two hours per charge from a removable 25.2 V battery and USB-C charging up to 140 W.

Infographic of the official Jumper hardware specifications including dimensions, weight, Rockchip RK3576, 22 servos and sensors

Open source, with a specific boundary

The project ships under Apache-2.0 and includes the MuJoCo simulation, the reinforcement-learning tasks, the PPO training code, GPU and CPU physics backends, the deployment path from policy export to the NPU, plus libraries of ready-made appearances and scenes. It also includes instructions so an AI coding assistant can drive the workflows from natural language.

What it does not include is a buildable robot. Full mechanical CAD, the production bill of materials, the PCB design and the servo manufacturing data are not published, and the project states that NPU inference on the real board has not yet been validated and the 1 kHz controller loop does not yet drive the real motor bus. In practice this is a software and AI development platform for an existing machine — not a plan for assembling one from scratch.

Infographic showing the three Jumper workflows, what is included in the open-source repository and what is not

On the move

Jumper crab robot standing on grass next to a football with the goal net behind it

Jumper crab robot on a desk carrying toy minifigures on a small brick platform

Jumper crab robot wearing a lucky cat shell with cat face graphics on a carpet

Jumper crab robot wearing a yellow bee shell on a leather sofa

A child playing on the floor with the Jumper crab robot next to wooden blocks, with an open source badge

Change the shell, build a scene

Three Jumper shell designs rendered in the design tool: warm sand ranger, silver armour and Raphael turtle

Three simulated Jumper scenes rendered in the design tool: a park pump track, a bedroom and a soccer pitch

A second repository handles appearance and scene design, exporting .skin and .map files, while trained movements are packaged as .app bundles. Official project links: the main repository, the design repository and the Crab Robot Open Source Challenge, which awards 100 engineering units to the best community work. Product page: kingkong.tech/jumper.