Segway has been a popular production as an alternative transporter since its invention at the end of 20th century. Millions of people like for its ergonomic design and high-tech elements. It is predicted to be an innovational product to change a person's life style. This thesis focuses on building a simple low cost, home-made Segway style scooter. This project uses two electric scooter motors, two 12V car batteries, one accelerometer and several microprocessors to build up the whole system. Significantly, this project also explains how to build a Brushed Direct Current (BDC) motor driver with a rated output power of more than 350W and the capability of coping with up to 120A transient peak current and up to 40A continuous current. Four-quadrant operation and eight modes of DC motor operation are discussed. A mathematical model of the Segway style scooter is also introduced in details. This including the modelling of a BDC motor, a two-wheeled inverted pendulum and their combination. The linearization of these models is used. At the end the linearized model is simulated in computer software.

There is a need for smaller and more economic transportation systems. Two-wheeled inverted-pendulum machines, such as the Segway, have been proposed to address this need. However, the Segway places the operator on top of a naturally unstable platform that is stabilized by means of a control system. The control stability of the Segway can be severely affected when minor disturbances or unanticipated conditions arise. In this thesis, a dynamic model of a Segway is developed and used in simulations of various conditions that can arise during normal use. The dynamic model of a general two-wheeled inverted pendulum and human rider is presented. Initial estimates of the parameters were calculated or obtained from other references. The results from numerous experiments are presented and used to develop a better understanding of the dynamics of the vehicle. The experimental data was then used to adjust the model parameters to match the dynamics of a real Segway Human Transporter. Finally, the model was used to simulate various failure conditions. The simulations provide a better understanding of how these conditions arise, and help identify which parameters play an important role in their outcome.

This thesis presents the design, construction, and control of a two-wheeled inverted pendulum (TWIP) robotic walker prototype for assisting mobility-impaired users with balance and fall prevention. A conceptual model of the robotic walker is developed and used to illustrate the purpose of this study. A linearized mathematical model of the two-wheeled system is derived using Newtonian mechanics. A control strategy consisting of a decoupled LQR controller and three state variable controllers is developed to stabilize the platform and regulate its behavior with robust disturbance rejection performance. Simulation results reveal that the LQR controller is capable of stabilizing the platform and rejecting external disturbances while the state variable controllers simultaneously regulate the system’s position with smooth and minimum jerk control. A prototype for the two-wheeled system is fabricated and assembled followed by the implementation and tuning of the control algorithms responsible for stabilizing the prototype and regulating its position with optimal performance. Several experiments are conducted, confirming the ability of the decoupled LQR controller to robustly balance the platform while the state variable controllers regulate the platform’s position with smooth and minimum jerk control.

Reinforcement learning is conceptually based on an agent learning through interaction with its environment. This trial-and-error learning method makes the process prone to situations in which the agent is stuck in a dead-end, from which it cannot keep learning. This thesis studies a method to diminish the risk that a wheeled inverted pendulum, or WIP, falls over during training by having a Qlearning based agent estimate a PID controller before training it on the balance problem. We show that our approach is equally stable compared to a Q-learning agent without estimation training, while having the WIP falling over less than half the number of times during training. Both agents succeeds in balancing the WIP for a full hour in repeated tests. / Reinforcement learning baseras på en agent som lär sig genom att interagera med sin omgivning. Denna inlärningsmetod kan göra att agenten hamnar i situationer där den fastnar och inte kan fortsätta träningen. I denna examensuppsats utforskas en metod för att minska risken att en självkörande robot faller under inlärning. Detta görs genom att en Q-learning agent tränas till att estimera en PID kontroller innan den tränar på balanseringsproblemet. Vi visar att vår metod är likvärdigt stabil jämfört med en Q-learning agent utan estimeringsträning. Under träning faller roboten färre än hälften så många gånger när den kontrolleras av vår metod. Båda agenterna lyckas balansera roboten under en hel timme.