In the field of humanoid robots and embodied intelligence, UC Berkeley is undoubtedly a top player. Last year, their team created the powerful humanoid robot Berkeley Humanoid for under $10,000, attracting widespread attention in the industry.
Recently, they have made another breakthrough— with just a 3D printer, you can replicate a Berkeley Humanoid Lite (the simplified version of the Berkeley humanoid robot) at home:
All hardware and software code are open source. You can also DIY your own robot.
This robot is impressive; it can solve a Rubik’s cube and unpack deliveries:
It can even write its own initials:
▍Technical Breakthrough: 3D Printing Solution Achieves Low-Cost Humanoid Robot
How cheap is this robot? It is said that sourcing all components in China costs about $3,236 (approximately 23,000 RMB), while in the US it is around $4,312, keeping the total cost under $5,000.
In contrast, products from commercial robotics companies like Boston Dynamics and Agility Robotics have never been priced below $100,000, often reaching hundreds of thousands of dollars. Of course, commercial versions have unique advantages in performance, stability, and industrial applications, but this also means that most researchers and makers find it difficult to access such platforms.
This robot stands about 0.8 meters tall, roughly equivalent to a toddler’s height, and weighs 16 kilograms. It is equipped with two 5-degree-of-freedom robotic arms and integrated grippers, as well as a bipedal walking system, fully capable of meeting walking and basic operational tasks.
The research team has fully utilized the latest advancements in desktop 3D printing technology. All non-standard structural components are manufactured using Fused Deposition Modeling (FDM) technology, significantly lowering the manufacturing threshold.
The actuator housings, cycloidal gears, input shafts, and output shafts are all 3D printed, with the input shaft embedded in a brass hex bracket to enhance rigidity and improve torque transfer from the motor to the cycloidal disk.
The team designed a new 3D printed cycloidal gearbox for the robot’s 22 joint actuators. Why choose a cycloidal gearbox? Because cycloidal gears can distribute force across multiple teeth, effectively avoiding layer adhesion failure—this type of failure is common in 3D printed structures after PLA material supports break.
This design not only reduces costs but also improves durability, making the robot easier to maintain and upgrade. The team conducted a continuous operation test lasting 60 hours, proving that this 3D printed gearbox can operate stably, and even with slight wear, it does not affect overall performance.
The Berkeley team also cleverly avoided the pitfalls of proprietary components. Many open-source robot designs often rely on hard-to-source custom thrust bearings or high-performance motor drivers. In contrast, Berkeley Humanoid Lite primarily uses common parts that can be easily purchased on platforms like Amazon and Taobao, while electronic components come from reputable suppliers like Digikey and Mouser.
By selecting parts that are widely available in multiple countries, Berkeley Humanoid Lite minimizes supply barriers, ensuring users can quickly obtain the necessary materials without geographical restrictions.
The core of the system consists of an Intel N95 mini PC and a 4000 mAh battery, providing a maximum runtime of about 30 minutes. The entire robot features a streamlined design, with all non-standard parts designed to fit within a 200×200×200 mm print volume, allowing for printing with a common FDM desktop 3D printer.
▍Simulation to Reality! AI Algorithms Seamlessly Transfer to Physical Robots
In laboratory tests, Berkeley Humanoid Lite demonstrated impressive capabilities, comparable to expensive commercial products.
Most notably, a reinforcement learning controller trained entirely in a simulated environment was able to start and walk on the real robot without manual adjustments. This is a coveted “zero-shot” transfer milestone in the field of bipedal AI, indicating that the robot’s physical characteristics are highly consistent with the simulation model.
Actual demonstration videos showcase the robot’s ability to perform a series of complex tasks, including:
● Writing initials
● Manipulating blocks
● Solving a Rubik’s cube
● Unpacking and restoring
● Walking smoothly at a specified speed
The research team specifically designed an evaluation metric to compare the cost-effectiveness of different robots: the average peak joint torque divided by the robot’s size, then normalized by price. According to this standard, this robot outperformed several commercial platforms priced an order of magnitude higher.
Even more surprisingly, the team developed a remote operation device based on the SteamVR tracking system, creating an intuitive operating interface. By wearing VR equipment, operators can control the robot to complete various tasks from any direction using handheld controllers, which is very intuitive.
▍Open Source Without Reservations! From 3D Models to Code, Build What You Want
Most importantly, all CAD files, firmware programs, and training scripts required for the project have been released for free under an open-source license on GitHub. The project address is:
HybridRobotics/Berkeley-Humanoid-Lite.
This means anyone can follow the detailed guide provided by the project to assemble and improve this robot. Most off-the-shelf parts can be shipped within a week in the US and China, while custom parts can be printed within a week, allowing the entire robot to be assembled in about three days.
If you want to assemble a Humanoid Lite yourself, you will need:
● 10 high-torque 6512 actuators ($188 each)
● 12 lighter 5010 actuators ($136 each)
● An Intel N95 mini computer
● A 6-cell 4,000 mAh lithium polymer battery
● Other miscellaneous parts
The Berkeley team initiated this research because they recognized the high hardware costs, low transparency, and poor customization in the current humanoid robot field. They hope to lower the barriers to robot accessibility through this open-source project, making robotic technology more inclusive and widespread.
This democratized design is reminiscent of the impact Raspberry Pi had on the computer industry over a decade ago. Just as Raspberry Pi made programming accessible to everyone at a very low price, Berkeley Humanoid Lite is bringing humanoid robot technology out of the lab and into a broader audience.
This project is supported by the National Science Foundation (NSF) of the United States (grant numbers 2303735 and 2238346) and holds public shares in the Robotics and AI Research Institute. This official support further ensures the project’s sustainability and reliability.
The research team also integrated an anime-style head into the robot, allowing it to exhibit expressive and visually appealing anime character features, enhancing overall aesthetics and increasing human-robot interaction potential.
What do you think about building such a humanoid robot for 23,000 RMB?
Project Address:https://lite.berkeley-humanoid.org
Paper Download Address:https://lite.berkeley-humanoid.org/static/paper/demonstrating-berkeley-humanoid-lite.pdf
Source: Humanoid Robot Release
For inquiries regarding corporate cooperation, please contact Dr. Tang (13810423387, same number for mobile and WeChat) for coordination.
—————-END——————-
Industrial Robot Companies
Estun Automation | Aofei Robotics | Feixi Technology | Faao Robotics | Yuejiang Robotics | Jeka Robotics | Songling Robotics | Luoshi Robotics | Yinglian Technology | Youao Robotics | CGXi Changguangxi Intelligent Manufacturing | Atom Robot
Service and Special Robot Companies
Yijiahe | Jingpin Special Equipment | Ninebot Robotics | Qiteng Robotics | Pudu Robotics | Xiaoben Intelligent | Robot Ji
Medical Robot Companies
Yuanhua Intelligent | Tianzhihang | Sizherui Intelligent Medical | Jingfeng Medical | Tuo Dao Medical | Zhenyida | Shurui® Robotics | Luosenbot | Bangkec | Baihui Weikang | Dish Medical | Shuimu Dongfang
Humanoid Robot Companies
UBTECH Robotics | Yushu | Yundongchu | Xingdong Jiyuan | Weijing Robotics | Zhujidongli | Leju Robotics | Elephant Robotics | Zhongke Shengu | Magic Atom | Zhongqing Robotics | Pashini Perception | Cyborg Robotics | Digital Huaxia | Ligong Huahui | Fourier Intelligence | Tianlian Robotics | Kepler Humanoid Robotics | Lingbao CASBOT | Qingbao Robotics | Zhejiang Humanoid Robot Innovation Center
Embodied Intelligence Companies
Cross Dimension Intelligence | Galaxy General | Qianxun Intelligent | Lingxin Qiaoshou | Ruierman Intelligent | Weiyi Intelligent Manufacturing | Tuihang Technology | Zhongke Silicon Chronicle | Shutu Technology | Lingqiao Intelligent | Xingchen Intelligent | Qiongche Intelligent | Ark Infinite | Keda Xunfei | National and Local Jointly Build Embodied Intelligence Robot Innovation Center | Qingtian Intelligent | Daimeng Robotics
Core Component Companies
Green Harmonics | Yinshi Robotics | Kunwei Technology | Maitai Intelligent | Qingtong Vision | Benmo Technology | Xinjingcheng Sensors | BrainCo Strong Brain Technology | Yuli Instruments | Jiya Precision Machinery | Silan Technology | Shenyuan Sheng | Feipu Navigation Technology | Ruichi Zhiguang | NOKOV Measurement Technology | Yinxike | Landian Touch | Fude Robotics | Cancer Intelligent Drive | Weihan Power | Yuanke Vision| Xuanji Power
Educational Robot Companies
Silicon Step Robotics | Shihe Educational Robotics | Daran Robotics
Join the Community
Welcome to join the “Robot Lecture Hall” reader discussion group, to discuss topics related to robotics, share cutting-edge technology and industry dynamics, add WeChat “robospeak2018” to join the group!
Feeling tired? Click “See More” to support us!