Estimation and Planning for Dynamic Robot Behaviors
CitationVarin, Patrick. 2021. Estimation and Planning for Dynamic Robot Behaviors. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.
AbstractSince the conception of modern robotics it has been a vision of roboticists, science fiction authors, and technologists to see machines with superhuman abilities propel humanity into the future. Grand visions of a robotic future see robots conducting space exploration, augmenting physical human abil- ities, and solving some of the most challenging societal issues that face us today.
Today we see glimpses of this future with robots doing useful work in a variety of fields: working in collaboration with humans in automotive and aerospace manufacturing, accelerating medical research
by automating repetitive tasks in biotech labs, augmenting the abilities of surgeons by enabling super- human precision and superfine motor movements, assisting the physically disabled to walk again, and surveying remote and dangerous locations. We have developed many physically capable robots and robotic hardware continues to progress at amazing rates.
In many cases, the capabilities of modern robots are not limited by the hardware but by the software and control systems that run them. In order to understand the full potential of our current robots, and the future of robotics, we must be able to push the physical limits of these mechanisms and understand how to operate at the boundaries of their capabilities. To achieve this, robots need to be intimately aware of their own dynamics and the dynamics involved in their interactions with the world. We need to understand where our dynamic models begin to fail, and how to be robust to the type of model errors that arise at the boundaries of a robot’s capabilities. This understanding should be reflected throughout the software control architecture in planning, control, and state estimation.
In this thesis we explore techniques for supporting “highly dynamic” robot behaviors in planning and state estimation. We discuss techniques for state estimation that is aware of and robust to the types of model errors that begin to appear in the fast, powerful behaviors that push the boundaries of a robot’s capabilities. We also discuss techniques for designing the types of behaviors that allow us to extract the maximum performance from modern robots.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37370219
- FAS Theses and Dissertations