3D-Printed Open-Source Humanoid Robot: A Step Towards Accessible Robotics for All

As an undergraduate student, Yufen Chi (B.S.’23 EECS) was captivated by humanoid and walking robots. Eager to expand his knowledge, he watched YouTube videos and engaged in class projects, but gaining practical experience and working on his own proved challenging.

«I was keen on building my version of a humanoid robot, but at that time, components like actuators, motors, and robotics kits weren’t readily available, and developers kept their source code under wraps,» he explained. «So, I started exploring ways to make a robot from scratch.»

Now a graduate student in the Department of Electrical Engineering and Computer Sciences, Chi is part of a team of engineers from Berkeley who created Berkeley Humanoid Lite—a cost-effective, open-source robot made from 3D-printed parts. They recently presented their findings at the 2025 Robotics and Systems Conference.

Humanoid robotics has advanced rapidly in recent years, with devices being designed for industrial automation, healthcare, research, and personal assistance. Despite the increasing interest in this field, Chi noted that most commercially available devices remain relatively expensive.

Over time, the issue of affordability has become easier to tackle. However, Chi pointed out that dealing with personalization and resource management remains a more complex challenge.

«Commercial companies can offer fully assembled robots, but due to the proprietary nature of their hardware and software, it’s often difficult to replace parts or modify components. This limits the opportunities for novice roboticists to experiment with customization and further explore the boundaries of humanoid technology.»

At the same time, not everyone has access to cutting-edge equipment. «Some research projects assume you have access to advanced CNC machines for precision fabrication and that you’re using specialized PCBs for electronics,» Chi said. «In a fully equipped lab, building a robot from scratch might be feasible, but for many, including hobbyists and DIY enthusiasts, it’s simply out of reach.»

While developing Berkeley Humanoid Lite, the researchers aimed to create a starting point for anyone interested in humanoid technologies.

«Our goal is to help researchers and educators understand how humanoid systems function, including how to assemble and develop a robotic platform by providing them with a template to get started,» he stated. «It’s about studying an example system and building it from the ground up, piece by piece. As you gain experience and confidence, you can then enhance it and take it to the next level.»

At the heart of the Berkeley Humanoid Lite design is a modular 3D-printed gearbox for the actuators and robot casing. All other components can be purchased from widely accessible e-commerce platforms or produced using standard desktop 3D printers.

As a result, the total cost for the equipment does not exceed $5,000 (based on U.S. market prices), which is just a fraction of the cost of buying a comparable mass-produced robot.

Moreover, with a 3D printer, it’s easy to create replacement parts for broken or worn components.

Standing about 1 meter tall and weighing approximately 16 kg, Berkeley Humanoid Lite can be assembled by a beginner in roughly a week, though this duration may vary depending on individual skill levels and experience.

«The good news is that in our Discord and other community chats, we see users actually building it,» he said. «They share photos of their assembled robots, which is very exciting.»

Since Chi began working on Berkeley Humanoid Lite four years ago, new startups have started offering more affordable metal actuators. However, Chi believes that the modularity of Berkeley Humanoid Lite gives it a significant advantage over commercial products.

«With our approach, you can begin by creating a single actuator, getting it to rotate, and then try to add several actuators to a simple arm or leg,» he explained.

Understanding that 3D-printed parts inherently lack the durability of materials like aluminum, the researchers incorporated a cycloidal gear mechanism into the actuator’s gearbox.

«The main advantage is that the gear teeth are very large,» Chi said. «This distributes the load over a larger surface area than in traditional geared systems, reducing stress and wear.»

Additionally, they tested several features of the 3D-printed actuators to ensure their longevity. «Our results showed that the 3D-printed actuator is at least on par with other actuators,» Chi stated, «or operates within acceptable limits for performing these tasks and higher-level skills.»

He added, «We designed it so that if an actuator fails, you can simply print another gearbox and replace it. However, we haven’t yet broken any actuators on our test robots, even after all these experiments.»

The researchers also assessed Berkeley Humanoid Lite’s ability to perform basic tasks like grasping objects and moving forward.

To enable hand manipulation, the team constructed a teleoperation system for the robot. Using a joystick, they demonstrated Berkeley Humanoid Lite’s capability to pick up objects and interact with them, including solving a Rubik’s Cube. The researchers also applied reinforcement learning to create a locomotion controller that allows the bipedal robot to walk.

Chi noted that while the robot’s movement skills may be «a bit clumsy and not entirely graceful,» he anticipates that the Berkeley Humanoid Lite community will enhance the software and address any bugs over time.

The hardware, embedded code, and frameworks for training and deployment of Berkeley Humanoid Lite are all fully open source. The researchers wanted users to see how everything functions and easily customize the robot.

«I believe in the spirit of open-source communities, in the idea of an ecosystem where people share ideas and knowledge,» Chi expressed. «We hope that Berkeley Humanoid Lite will help bring us closer to democratizing humanoid robot development.»

In addition to Chi, the co-authors of the research are Kushil Srinath, an associate professor in the Department of Mechanical Engineering and a principal investigator; mechanical engineering graduate students Qiaoyuan Liao, Junfeng Long, Xiaoyu Huang, and Zhongyu Li; along with Sophia Shao, an associate professor, and Borivoje Nikolic, a professor, both from the Department of Electrical Engineering and Computer Sciences.