TROT
During my master's program in Michigan, I worked on developing a quadruped robot named TROT. TROT (The Robot of Theseus) is a fully 3D-printed, cost-effective, and open-source robot designed to explore the possibilities of leg morphology changes. It was built from scratch under the guidance of prof. Talia Moore at the Embir Lab. The Interesting thing about this robot is that it can change leg morphology (gonna explain what that is below). This unique challenge made it an incredibly fascinating project, and much of my time in Michigan was spent working on this robot.
Have you ever noticed that different quadruped animals, like dogs, cats, and horses, all have distinct leg morphologies? Specifically, the ratio between their tibia (the bone below the knee to the ankle) and femur (the bone above the knee to the hips) varies. In fact no animal has 4 legs that look identical. The front legs and back legs are always different. To figure out why, we first need to figure out what is the result when we change all four legs to different ratios, then test a morphology with the fore-hind limb differences. The main goal of building TROT was to explore how changing these leg proportions—between the tibia, femur, and other parts—affects performance. By experimenting with different leg morphologies, we aimed to identify which configurations are most suited for specific situations and gain a deeper understanding of how animal leg structure influences mobility and efficiency.
Hence TROT was born. We built new hardware, tweaked some software, and basically did whatever it took to make the bot work. Let me give you a brief overview of how this was done -
Any robot requires actuators to move, and choosing the right actuator is crucial, especially for a legged robot. Before I joined Michigan, some of my labmates were already working on selecting the optimal actuator. For a legged robot like TROT, we needed an actuator that is fast, lightweight, and capable of generating enough torque to lift the robot. Essentially, we were looking for an actuator with low mass, low moment of inertia, a high torque constant, and a high motor constant.
While most existing research suggested selecting actuators based on their thermal specific torque, the team I worked with proposed a different approach and developed a new metric for actuator selection. The actuators we designed were 3D-printed and performed just as well, if not better, than those used in the Mini Cheetah robot! I played a key role in building the actuators for TROT and conducting experiments that contributed to this research. You can read more about our findings here.
TROT was entirely 3D-printed and built using affordable, off-the-shelf components, which made it incredibly accessible and lowered the barrier to entry for building the robot. With just a 3D printer and access to a local hardware store, anyone could replicate TROT. Even the actuator's gearbox was 3D-printed! To transfer torque to the robot's legs, we used 4-bar linkages, which allowed us to position most of the actuator's mass on the robot's body. This design kept the legs lightweight and highly agile, giving TROT exceptional mobility and responsiveness.
We used the LCM as the communication protocol (we thought ROS had too much bloatware). The software was the Mini cheetah controller that we forked and modified to work with out hardware. It took us a lot of debugging of CAN-FD stuff and mini cheetah code to get the robot work, but hey it finally worked!
One of the main issues with the robot was that it drifted in the yaw axis, this is because we used a 6 axis IMU without a magnetometer. To fix this I worked on a sensor fusion for a 9 axis IMU.
TROT testing
Slow Mo of trot walking on a treadmill