Developing Novel Hardware Platforms for Model-Predictive Control

Abstract

Robot hardware has rapidly become more accessible in the last ten years.
However, there is still a dearth of low-cost hardware platform that are
open-source and easy to build. With recent developments in accessible
manufacturing methods, such as FDM 3D printing, manufacturing and
designing parts without using precision machining has become feasible.
In this thesis, we address two gaps in open-source hardware platforms
that can be filled using relatively few precision machined parts. Both
designs are fully open-source, and the controllers for both systems are
also available online.
The first hardware platform aims to create a lightweight bipedal system
with pitch control, allowing it to recover from perturbations and perform
dynamic motions. We do this by adding a reaction wheel actuation system
that controls the pitch angle of the robot, fully actuating the system and
enabling attitude stabilization. We linearize the dynamics of the system
to obtain a linear discrete-time optimization problem that can be solved
as a quadratic program. The linear MPC problem is tested on hardware
at 240 Hz.
The second part of the thesis covers the design and control process for an
aquatic robot that can be easily modeled in a fluid simulator, simplifying
the sim-to-real transfer process. To precisely control the robot state, we
use a five-bar linkage system. We implement and test this on hardware,
and demonstrate that our system is able to accurately and repeatably
navigate in a fluid environment.