Research

Current Projects

Custom Servo Project

Our humanoid robot has 18 degrees of freedom, each of which makes use of a smart servo motor. Considering that 4 robots have to be made to form a complete team, the cost of servos adds up very quickly. Rather than looking for funds and worrying about handling the off-the-shelf servos, we asked: is it possible to reduce the overall cost at the expense of taking on a greater technical challenge?

We saw an opportunity in answering this question, one that will provides students with learning experiences about electrical motors and their controls.

Semi-Custom Solution for Servo Motors

Since the risk of investing in a fully-custom solution that provides the same functionality as off-the-shelf serovs is high, we decided to consider a compromise between the two with a reasonably lower risk. By modifying and/or adding components to the cheaper hobbyist servo motors, we aim to provide the very basic functionality that currently our robot relies on: position control. We estimate the success rate of this project to be 80%.

Fully Custom Solution for Servo Motor

Our research for the custom servo project led us to believe brushless DC (BLDC) motors will provide power output at the same level or higher than the brushed DC motors used inside the off-the-shelf solutions while keeping the cost of the motor significantly lower and providing higher efficiency and reliability. However, all of these benefits come at the cost of the technical difficulty of operating the BLDCs. That is, depending on the location of the rotor (the moving part of the motor), the controller must change which coils pass current at any instant of time. Starting from principles of driving BLDCs, the goal of this project would be provide an abstract interface for torque control while keeping the costs low and physical dimensions reasonable.

Our current plan is to study the BLDC drive principles using the STEVAL-SPIN3201 evaluation board from STMicroelectronics to develop our custom solutions, and later use the STSPIN32F0 integrated solution in the final result. STSPIN32F0 brings together a BLDC motor driver and a powerful 32-bit ARM microcontroller, which makes it the ideal component to design our systems around.

Learn more about STSPIN32F0 in the video below:

Reinforcement Learning Infrastructure Project

This year, we are using the latest reinforcement learning (RL) tools to optimize our existing walking algorithms. Setting up the infrastructure for this is indeed a very ambitious project to undertake; we are hoping it can come close to the standards of some of the world’s current reinforcement learning platforms. We are looking for people with proven expertise and interest in machine learning and software infrastructure to help set up this grand plan for future generations. Look at our inquiries to see how you can help our team out.

Reinforcement Learning Software Status

Reinforcement Learning Infrastructure Status