Motor skill learning with local trajectory methods
Abstract/Contents
- Abstract
- Motor or sensorimotor skills are behaviors that require close coordination of motor control with feedback from the environment. This includes a wide range of human and animal behaviors, such as locomotion and manipulation. Constructing effective and generalizable motor skills is crucial for creating naturalistic, versatile, and effective virtual characters and robots. However, constructing such motor skills manually requires extensive engineering and, quite often, nontrivial insights into the structure of the behavior. For a robot or virtual character to reproduce a motor skill repertoire as wide as that of a human being, the required engineering effort would be staggering. A more scalable approach is to acquire motor skills autonomously, by combining concepts from optimal control with machine learning. In this thesis, I discuss several algorithms based on local trajectory methods that can be used to construct motor skills for walking, running, swimming, traversal of uneven terrain, and recovery from strong perturbations. I show how example demonstrations can be used to automatically learn the objective or goal of the skill, and how local trajectory methods can be used to train general-purpose controllers, represented by large neural networks, without the need for extensive manual engineering or domain knowledge about the task at hand.
Description
Type of resource | text |
---|---|
Form | electronic; electronic resource; remote |
Extent | 1 online resource. |
Publication date | 2014 |
Issuance | monographic |
Language | English |
Creators/Contributors
Associated with | Levine, Sergey |
---|---|
Associated with | Stanford University, Department of Computer Science. |
Primary advisor | Hanrahan, P. M. (Patrick Matthew) |
Primary advisor | Koltun, Vladlen, 1980- |
Thesis advisor | Hanrahan, P. M. (Patrick Matthew) |
Thesis advisor | Koltun, Vladlen, 1980- |
Thesis advisor | Liang, Percy |
Advisor | Liang, Percy |
Subjects
Genre | Theses |
---|
Bibliographic information
Statement of responsibility | Sergey Levine. |
---|---|
Note | Submitted to the Department of Computer Science. |
Thesis | Ph.D. Stanford University 2014 |
Location | electronic resource |
Access conditions
- Copyright
- © 2014 by Sergey Vladimir Levine
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...