The Mini Cheetah Robot

Back during my second year of undergrad I started scheming about building legged robots out of cheap hobby brushless motors.  Two years later as a senior, I finally made a first-pass at building some of the hardware - motor controllers, gearboxes, and a two-degree of freedom leg.  Things look promising, so I stayed at MIT to do a masters in Sangbae's lab continuing the project.  The Mini Cheetah robot is the result of that work:

Photo: Bryce Vickmark

First quadruped robot to do a backflip!

The first version of the robot was put together in March 2018.  I spent the summer after finishing my Masters helping get the robot fully up and running (literally).  When we saw how well everything worked, we decided to build a handful more of the robots, both for our lab and to loan out to other research groups to experiment on.  I spent about a year doing some design revisions and building 10 more of the robots.

The robot is probably best documented in my masters thesis, but there are also several blog-posts as well.

Here are several papers (in no particular order) published about the robot or  using it.  I'll keep updating this list as I hear about more publications using the robot

• Mini cheetah: A platform for pushing the limits of dynamic quadruped control
• Benjamin Katz, Jared Di Carlo, Sangbae Kim
• Highly Dynamic Quadruped Locomotion via Whole-Body Impulse Control and Model Predictive Control
• Donghyun Kim, Jared Di Carlo, Benjamin Katz, Gerardo Bledt, Sangbae Kim
• Vision aided dynamic exploration of unstructured terrain with a small-scale quadruped robot
• D Kim, D Carballo, J Di Carlo, B Katz, G Bledt, B Lim, S Kim
• Hybrid Systems Differential Dynamic Programming for Whole-Body Motion Planning of Legged Robots
• He Li, Patrick M. Wensing
• Model Hierarchy Predictive Control of Robotic Systems
• He Li, Robert J. Frei, Patrick M. Wensing
• Regularized predictive control framework for robust dynamic legged locomotion
• Bledt, Gerardo.
• Extracting Legged Locomotion Heuristics with Regularized Predictive Control
• Gerardo Bledt; Sangbae Kim
• Variational-Based Optimal Control of Underactuated Balancing for Dynamic Quadrupeds
• M Chignoli, PM Wensing 
•  Robust Autonomous Navigation of a Small-Scale Quadruped Robot in Real-World Environments. 
• Dudzik, Thomas & Chignoli, Matthew & Bledt, Gerardo & Lim, Bryan & Miller, Adam & Kim, Donghyun & Kim, Sangbae. 
• Concurrent Training of a Control Policy and a State Estimator for Dynamic and Robust Legged Locomotion
• Gwanghyeon Ji, Juhyeok Mun, Hyeongjun Kim, Jemin Hwangbo
• Learning to Jump from Pixels
• Gabriel B. Margolis, Tao Chen, Kartik Paigwar, Xiang Fu, Donghyun Kim, Sangbae Kim, Pulkit Agrawa
• Agile Locomotion via Model-free Learning
• Gabriel Margolis*, Ge Yang*, Kartik Paigwar, Tao Chen, Pulkit Agrawal

The Rubik's Contraption

Jared Di Carlo and I built a machine that can solve a Rubik's cube really fast, and it became kind of internet-famous:

See my hardware and low-level control build-log here

and Jared's software blog here

Furuta Pendulums

In the summer of 2016 I built a desktop furuta pendulum out of random parts I had lying around:

Recently, I redesigned it to not be built out of scraps, sent out for all the machined parts, and assembled a handful more:

See the build logs here

Motor Dynamometer

In 2016 I built an electric motor dynamometer - a device for characterizing the performance of electric motors and motor controllers



The dynamometer is controlled by a Python and QT-based interface which provides manual control of the torque, speed, voltage, and current, simulates loads, or runs through sequences of test points.
The dyno lets me characterize electric motors to generate useful data like efficiency maps:



See the build log here:

Tree Climbing Robot

After I got comfortable programming and building with an Arduino, I used my newly acquire microcontroller skills to build a robot.  Using a microcontroller, four high-torque DC gear motors, spiked legs, a linear actuator, rotation sensors, and 3 L298 H-bridge circuits, this robot can climb up trees of varying diameter.  The very long build log follows.



The robot in action: