AnswersPublished29 sources
KingKong’s Jumper Robot: What’s Open Source, and What Still Needs Testing
Jumper is a six legged, 22 DoF robot paired with software for training and packaging motions. The repository was public by September 28, 2026, so October 9 was not its first public appearance.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Jumper is a six legged, 22 DoF robot paired with software for training and packaging motions.
- The repository was public by September 28, 2026, so October 9 was not its first public appearance.
- The toolkit supports CPU and GPU simulation workflows and exports policies toward the RK3576 NPU; that is not the same as proving a complete, reproducible robot build.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Beijing-based encoder and servo maker KingKong Technology reveal in its October 9, 2026 open-source release of Jumper, including th. Article summary: KingKong’s Jumper release opened a robot-motion development toolkit—simulation, training, policy export and controller packaging—rather than establishing a fully reproducible open-hardware robot or a validated end-to-end. Topic tags: general, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks,
KingKong Technology’s Jumper is both a compact crab-style robot and a software toolkit for creating its motions. The project brings together robot models, simulation, reinforcement-learning training and controller packaging. But an open software workflow is not proof of a fully reproducible hardware design or a validated deployment on the physical robot. The repository was public by September 28, 2026, and coverage followed before October 9, so that date should not be treated as the initial release date.3
9
Jumper’s design and published specifications
Jumper has six legs and 22 degrees of freedom, driven by 22 tactile servos. Its front appendages can grasp objects.6
14 The hardware specification lists a body measuring 400 × 400 × 200 mm and a weight of 1.8 kg; it also notes that an engineering prototype weighs 2.8 kg.
21
KingKong’s published materials give a top speed of at least 0.5 m/s and about two hours of battery life. A report also lists a jump height of at least 40 cm and a grasping capacity of 1 kg. These are stated specifications, not independently verified results in the available evidence.14
16
The robot’s main computer is a Rockchip RK3576, with four Cortex-A72 cores, four Cortex-A53 cores and a 6 TOPS INT8 NPU.21 A report describes two magnetic encoders in each tactile servo, but the repository issue tracker notes that the servo model or type is not specified in its documentation.
27
40
What the open-source toolkit lets developers do
The repository combines robot-model assets, tasks, training code and deployment tools. Developers can train motions, replay and evaluate policies, then export them in motion bundles.9
11
24
The training stack uses MuJoCo, MuJoCo Warp, mjlab and rsl_rl. Project documentation says the same task configurations, rewards and PPO training code can run with either MuJoCo Warp for GPU simulation or native MuJoCo on CPU.18
The official tutorial walks through training tripod-gait, jump and dance policies and combining them in one bundle.19 The broader workflow is designed to let developers modify and retrain motions rather than rely only on fixed demonstrations. The project also invites people to use an AI coding assistant to request motion, appearance or scene changes in natural language. Reported outputs include
.app motion bundles, .skin appearances and .map scenes.5
9 That authoring approach is an intended workflow, not evidence that arbitrary prompts will reliably produce safe real-world behaviors.
How policy export and deployment are intended to work
Policies can be exported as ONNX models and converted to RKNN for the RK3576 NPU. A motion bundle can package policies for different modes alongside a finite-state-machine configuration, controller builds for supported hosts and operating instructions. The documented hosts include the robot board, a browser and simulation or replay tools.19
32
38
The key caveat is validation. Project documentation describes training, simulation, export and host-side bundle checks as implemented, but says inference on the robot’s RKNN hardware had not been verified. The controller also had not been shown driving the live motor bus; the available evidence does not establish that the documented 1 kHz control loop was running on the physical robot.18
32
41
Open software is not the same as a complete rebuild package
Jumper’s release is best understood as an open motion-development toolkit, not as confirmation that anyone can reproduce the robot from the published materials alone. The available sources do not establish that complete mechanical, electrical and manufacturing documentation was released. In particular, the repository issue tracker’s note that servo types are unspecified illustrates a gap in the build information.40
That distinction matters: developers can work with the supplied models and software without having all the information needed to manufacture identical hardware. Claims that every part needed for a faithful rebuild is available should therefore be treated cautiously.
Why the release matters
KingKong is known for encoders and servos, and Jumper marks a move toward presenting its own robot platform alongside the tools for creating its motions.7
15 The project also connects simulation-based development with a physical robot target, while making the boundary between those stages visible: policy training and packaging are documented, but live onboard execution remained unverified in the cited materials.
18
32
A developer motion challenge is part of the project’s ecosystem, and a public community entry demonstrates trained Jumper motions in a game setting.14
30 That is evidence of creative experimentation, not proof of established applications in household chores, inspection, sorting or farming. Those remain possible directions rather than validated deployments.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "KingKong’s Jumper Robot: What’s Open Source, and What Still Needs Testing"?
Jumper is a six legged, 22 DoF robot paired with software for training and packaging motions.
What are the key points to validate first?
Jumper is a six legged, 22 DoF robot paired with software for training and packaging motions. The repository was public by September 28, 2026, so October 9 was not its first public appearance.
What should I do next in practice?
The toolkit supports CPU and GPU simulation workflows and exports policies toward the RK3576 NPU; that is not the same as proving a complete, reproducible robot build.