AbstractThe use of robotic legged locomotion offers great potential in terms of the ability to traverse discontinuous or rough terrain, inaccessible to wheeled or tracked robots, because feet require only intermittent support. In order to cross a complex environment with limited safe foothold positions, a legged robot must be able to accurately control the length of each step independently. For a hopping or running robot, this requires the precise control of both apex height and forward velocity, with demand values varying on every step – an agile gait – which is the challenge this thesis contributes towards.
In particular, the transition between the apex of one step and the next is considered as a discrete problem of selecting the system inputs which result in the desired next apex state, consisting of height and velocity. Despite the near ubiquity of legged locomotion in nature, this task has proved difficult in robotic control. This is due to the need for highly dynamic gaits, away from the simplifying assumptions of static balance, and the use of spring-like mechanics to store and re-use energy between steps. The Spring Loaded Inverted Pendulum model is widely used as an intuitive and simple model which captures the key characteristics of dynamic running and hopping, but even this lacks an exact analytical solution making the control task difficult.
The approach taken in this work is to formulate a simple controller, based on simple analytical analysis, but select the gain values within this controller by using the results of previous steps, either pre-computed or on-line based on the preceding steps. It is shown to be as accurate or better than the available analytical approximations from the literature when applied to the Spring Loaded Inverted Pendulum model, whilst remaining computationally inexpensive -- an important advantage over numerical methods for real-time implementation. Furthermore, by tuning the controller on-line based on previous steps, this approach has the benefits of reducing the manual effort in implementation and automatically adapting to changes in the system dynamics during a robot's operation.
The control tasks for apex height and forward velocity are first considered independently, with the proposed approach applied to each in simulation and using physical experimental hopping robots, before finally they are combined to control both aspects simultaneously in a simulation model.
|Date of Award||24 Jun 2020|
|Supervisor||Pejman Iravani (Supervisor) & Jonathan Du Bois (Supervisor)|