on april 26, embodied intelligence company kinetix ai (kai) held the “gifted arrival” launch event, officially unveiling its namesake first full-size humanoid robot, “kai.” this launch differed from typical product launches: rather than being introduced in a conventional manner, kai made its debut through a “dialogue and self-introduction” featuring two kai units, who described their capabilities, technological roadmap, and product positioning.
co-founder tyler explained at the event: “for a robot to truly integrate into the human world, it must be more anthropomorphic. he believes that human physical intelligence arises from interaction with the human environment. this requires robots to achieve closed-loop capabilities in at least three key areas—understanding the world, learning from the world, and interacting with the world—which correspond to a physical-world model, first-person datasets, and a highly anthropomorphic body.”
understanding the world means enabling the robot not only to perceive its current environment but also to predict future environmental changes—this is what kai’s world model accomplishes.
according to the presentation, superdimensional dynamics’ world model system has now achieved a closed-loop architecture. in addition to the core model, the system includes an action module and an evaluation module. the action module generates candidate actions based on the current state and passes them to the base model; the base model then predicts future states based on these candidates; and the evaluation module assesses the value of each trajectory, considering factors such as task progress and whether contact is safe.
“hi, i’m kai”—superdimensional dynamics unveils the world’s most highly degrees-of-freedom humanoid robot
while the world model addresses understanding the world, the first-person dataset tackles learning from the world. kai’s approach is to observe and engage with the world through human perspectives and movements, thereby acquiring experiences that are closer to real-world scenarios.
superdimensional dynamics has independently developed the scalable data-collection terminal kai halo. this device adopts a lightweight headband form factor with an “one-chip, eight-camera” configuration, enabling one-stop collection of human-world data. it can not only record first-person perspective video but also reconstruct human poses and scene point clouds. because the device is sufficiently lightweight, users can wear it naturally during daily activities, meaning the data-collection process does not require scripted choreography and the resulting data better reflects real-world conditions. compared with deliberately designed motion sequences, this type of data often captures a wider variety of movements, making it more suitable for training humanoid robots.
the world model allows kai to understand the world, and the first-person dataset gives kai “human experience,” but a robust training paradigm is still needed to transform that experience into skills—this is the three-stage training system.
during the pre-training phase, kai utilizes vast amounts of internet and simulation data, supplemented by self-collected first-person perspective data to enrich semantic information about whole-body and environmental interactions, thereby building a “common-sense” framework that aligns “space-language-vision-task actions.” in the bridging training phase, data from umi and data-gathering gloves are used to fill gaps in fine hand movements and physical contact that were underrepresented in the pre-training stage. in the post-training phase, real-machine remote-control data from specific scenarios are introduced to address alignment issues with the robot’s own body.
with data and a “brain,” the final step is to provide a capable physical platform—the highly anthropomorphic body—so that the robot can seamlessly integrate into everyday human life and interact safely with people.
this is kai’s body—kaibot. according to the presentation, kai’s highly anthropomorphic capabilities can be summarized across four dimensions: anthropomorphic form, anthropomorphic physique, anthropomorphic physical performance, and anthropomorphic perception.
in terms of form, kaibot stands 173 centimeters tall, weighs 70 kilograms, and has a head-to-body ratio of 1:8.5. its overall size and mass distribution closely resemble those of an adult human, ensuring stability while maintaining a human-like appearance and center-of-gravity placement.
in terms of physique, kaibot boasts 115 degrees of freedom, enabling shoulder elevation within a range of -20° to 0°, shoulder circumduction from -15° to 0°, neck movement from -15° to 50°, and lumbar movement from -15° to 75°—covering nearly the full range of human motion and more closely mirroring how human limbs operate naturally in complex environments. at the same time, kaibot is equipped with a dexterous hand featuring 36 degrees of freedom per hand, including 22 primary control degrees and 14 compliant degrees, allowing it not only to perform delicate tasks like grasping and pinching but also to absorb impacts, delivering safer and more natural interaction characteristics.
in terms of physical performance, kaibot is powered by a 1.7 kwh semi-solid-state battery, supporting approximately three hours of dual-arm operation. meanwhile, its custom-designed compliant actuators not only help the robot achieve motion smoothness closer to that of a human but also enable the arms to carry nearly 20 kilograms, balancing user-friendliness with operational capability.
in terms of sensory perception, kai is equipped with a full-body tactile skin system containing 18,000 touch sensors, theoretically capable of detecting even the slightest touches greater than 0.1 n and providing feedback. this tactile system enables the humanoid robot to collaborate closely and accompany humans in real-world settings, establishing a higher-quality feedback loop.
at the end of the press conference, kai stated: “i was not created to lift heavy objects in a factory with no human workers, nor was i designed to perform extreme stunts on stage. i was created to step into your living room or office, to understand your intentions, respond to your touch, and ultimately collaborate and coexist with you.”
