Unique robot built from off-the-shelf parts walks on two legs

A new open-source bipedal robot, dubbed MEVITA, has been designed by engineers at the JSK Robotics Laboratory of the University of Tokyo, Japan. 

Interestingly, MEVITA aims to tackle the common issues of existing DIY platforms. 

It’s designed to be strong and easy to build, using a minimal number of parts that are all available online, which makes advanced robotics more accessible to everyone.

18 metal components

There’s a growing trend to release bipedal robots as open-source platforms. This allows more people to develop these robots and share their knowledge without having to rely on commercial products.

However, the open-source robotics community has grappled with trade-offs. 

Many accessible robotics platforms use 3D-printed parts, which often results in fragile designs that can’t be easily scaled up.

On the other hand, while metal-based designs are durable, they often have so many components that they’re hard to assemble, and the parts aren’t easy to find online.

“To address these issues, we developed MEVITA, an open-source bipedal robot that can be built entirely from components available via e-commerce,” the researchers wrote in the study paper. 

The team used sheet metal welding to achieve the most basic functional design for a bipedal robot. 

This technique allows for the combination of complex shapes into a single piece, which drastically cuts down on the number of individual components needed. As a result, the robot is much easier for anyone to assemble.

The design of MEVITA focuses on a minimal and integrated structure. 

Excluding mirrored parts, the robot is built from just 18 unique metal components. A key feature is that four of these components are created using sheet metal welding. 

Advanced design

The brain behind this bidepal robot is equally streamlined. 

MEVITA’s control system uses modern AI techniques to achieve top performance.

It begins with control strategies trained through reinforcement learning in a simulation environment called IsaacGym. 

These trained behaviors are then verified in a different simulation, MuJoCo, before being deployed to the physical robot’s hardware using standard Python scripts. 

This Sim-to-Real transfer method proved highly effective, as demonstrated by the robot’s ability to walk across various challenging environments, including uneven indoor terrain, grassy fields, dirt surfaces, concrete tiles, and gentle slopes.

The circuit configuration for the bipedal robot is designed for efficient control and safety. 

The study noted “servo motors are connected to the PC via two CAN-USB interfaces, and the system is equipped with a wireless emergency stop, power relay, diode, and LiDAR/IMU.”

For safety, the system includes a wireless emergency stop, allowing for immediate shutdown from a distance. 

Moreover, the power relay and a diode are integrated to manage power flow and protect the system. 

And it is equipped with a LiDAR and an IMU (Inertial Measurement Unit) to provide crucial sensor data for navigation and balance.

The project aims to lower the barrier for bipedal robot development and encourage new research by making the entire platform—including its hardware, software, and training environments—open-source.

It could be used by researchers, students, and hobbyists, allowing them to build, innovate, and shape the future of bipedal robotics.