NEW $10999 Open Source Humanoid Robot Tech Breakdown (AI HARDWARE PRICE)

Kscale Labs is selling an open-source humanoid robot, Kbot, for $11,000. The robot has a modular head and body, allowing for customization and upgrades. It uses standard interfaces and 3D printed parts for easy modification. The robot’s design includes a clamshell chassis for durability and easy maintenance, modular hands, and hip joints that allow for natural movement. It’s powered by a lithium-ion battery that lasts for about four hours. All of Kbot’s hardware and software are open-source, making it a developer-friendly platform.

Kscale Labs just revealed what’s inside its open source Kbot humanoid that’s now selling for $11,000 come this December, so here’s the full breakdown of all the hardware inside, starting with the Kbot’s latest feature, its modular head. This design makes it easier for users to swap or upgrade the head to fit various applications, such as adding custom sensors or cameras. And for head attachment, Kbot uses a standard interface that can be replaced or updated with a 3D printed alternative for total customization. Next, inside its torso, two main printed circuit boards manage power distribution and actuator control, with the power board channels carrying 48 volts to each limb, providing communications daisy-chained from one actuator to the next.

Then, a separate controller area network board coordinates actuator signals through a simple USB connection to the head. In fact, future versions are expected to combine both boards into a single, more compact assembly to streamline internal wiring and enhance serviceability. And Kbot’s emergency stop functionality is also set for improvement, with integrated wired and wireless options coming soon. But for arms and legs, Kscale Labs adopted a clamshell chassis design, integrating structural support directly into the external shell, rather than using a traditional skeleton and cover approach. This improves impact resistance and simplifies maintenance by providing direct access to the robot’s wiring and motors.

It also features built-in hard stops with precise joint limits, which are critical for the robot’s safety and the accuracy of simulation models. As for hands, Kbot’s end effectors are also modular. While the current design offers a parallel gripper based on a compact four-bar linkage, the company is also evaluating five-fingered hands for the future. For now, though, the gripper attaches via a micro four-thirds camera mount for maximum stability and ease of replacement. But what’s more impressive is Kbot’s hip joints, which feature three degrees of freedom with a flexion-obduction rotation configuration, allowing for a flatback profile so the robot can lay or sit naturally.

And key to this are Kscale’s chosen actuators, which are quasi-direct drive planetary gear motors rated at up to 120 Nm of torque. And even more importantly, these units are off-the-shelf, mass-produced items to help keep costs and maintenance manageable while providing the necessary torque and responsiveness for agile humanoid movements. As for its lower limbs, they employ a parallelogram linkage at the ankle, with the motor mass positioned higher up the leg to reduce overall inertia. And to top it all off, the robot’s feet have been redesigned from flat to a rounded capsule profile, facilitating more realistic walking dynamics in both real-life and virtual training environments.

And to power it all, Kbot features a 13-ampere-hour lithium-ion battery pack, providing approximately four hours of continuous operation on a single charge. In fact, Kscale Labs has evolved its battery design from being an external handle to a completely flush-mount system on the robot’s back. Then, for wiring, in order to balance cost and ease of repair, the team opted against using hollow shaft actuators, instead, routing wires carefully around joints and bearings to minimize stress during movement. Amazingly, every aspect of Kbot’s hardware and software is open source, reflecting Kscale Labs’ commitment to developer-friendly robotics platform. And pre-orders for the Kbot Developer Kit are already open, with further deliveries scheduled for December of this year.

But tell us in the comments if you think $11,000 for this completely open-source, customizable humanoid is a fair price. Meanwhile, in China, to advance humanoid locomotion across complex terrains, researchers have introduced a novel framework called MOAR, or Mixture of Residual Experts, using Unitree’s G1 robot. The problem is that while reinforcement learning-based methods have helped robots achieve robust walking, most prior techniques are limited to flat surfaces and rely solely on proprioception, meaning they lack environmental awareness. But MOAR finally addresses this issue with the use of exteroceptive sensors, employing depth cameras to help the robot perceive and adapt to its surroundings.

Notably, the training actually occurs in two stages where first, the robot learns to navigate complex terrains, then, it gains the ability to switch between different human-like gate patterns through specific command inputs. On top of this, the framework introduces gate rewards for precise adjustments. The result is that in both simulation and real-world tests, MOAR delivers strong performance, while smoothly transitioning between multiple gate styles across challenging surfaces. Overall, MOAR is a critical step toward giving humanoid robots improved navigation in the real world. But robots are just one face of the AI revolution, as behind the machines, a quieter transformation is unfolding with Adobe having just made it easier than ever for creators to produce engaging video content, thanks to two new generative AI features, Text to Avatar and Generate Sound Effects.

With Text to Avatar, users can turn written scripts into captivating, Avatar-led videos in just a few clicks. This feature offers users a diverse selection of virtual presenters, plus the ability to customize backgrounds with a color, image, or video. What’s more is that creators can also select from a range of regional accents, ensuring the final product resonates with a global audience. And Firefly now handles all the heavy lifting by generating a fully polished video with a virtual host that’s ready for sharing or further editing. Because of this, Text to Avatar is already being recognized by educators, marketers, and businesses, with some use cases including producing clear video lessons or frequently asked questions with virtual presenters, transforming blog articles into scalable social media videos, or developing internal training materials with a personal touch.

Furthermore, the tool is designed for consistency so that teams can easily maintain a cohesive brand voice across large volumes of content, and Adobe’s Generate Sound Effects introduces an intuitive way to add custom audio to videos. Users can simply type a prompt, and Firefly’s artificial intelligence instantly produces the requested sound effect. For those seeking even more control, voice-guided input allows creators to dictate the timing and intensity of sounds, with Firefly matching the audio to on-screen action for cinematic timing. And both of these new tools are integrated with Adobe Express and Premiere Pro, meaning completed videos and soundscapes can be exported directly for immediate use or further refinement.

With these announcements, Adobe is pushing to compete in the generative AI video niche for the next evolution of social media. And finally, Apple has taken a significant step forward in accessibility technology with a new demonstration of brain-controlled features for iPhone and iPad devices. Earlier this year, the company announced its intention to support brain machine interfaces for users with compatible neural implants, and now this vision is becoming a reality. In a recently released video, Synchron, a leader in brain-computer interface technology, showcased Jackson, a participant living with amyotrophic lateral sclerosis, using only his thoughts to navigate an iPad’s home screen, open applications, and compose messages.

This live demonstration builds on top of Synchron’s StentRoad system, which uses an implanted network of electrodes to interpret brain signals into precise device commands. As of now, StentRoad has been implanted in 10 people as part of an FDA-approved trial, so tell us in the comments if you’d buy a DIY brain-computer interface implant kit, like and subscribe, and check out this video here. [tr:trw].