Page 84 |
Save page Remove page | Previous | 84 of 151 | Next |
|
small (250x250 max)
medium (500x500 max)
Large (1000x1000 max)
Extra Large
large ( > 500x500)
Full Resolution
All (PDF)
|
This page
All
|
S2R with model parameter estimation. A substantial body of work considers closing the
S2R gap by carefully estimating parameters of the real system to achieve a more realistic simulation.
For example, Lowrey et al. [2018] transfer a non-prehensile manipulation policy for a system of
three one-finger robot arms pushing a single cylinder. Tan et al. [2018] show transferability of agile
locomotion gaits for quadruped robots. Antonova et al. [2017] learn a robust policy for rotating
an object to a desired angle. While careful parameter estimation can provide a good estimate of
the model, it often requires sophisticated setups [F¨orster, 2015]. In contrast, we transfer a learned
policy to novel quadrotors for which we do not perform accurate model parameter estimation.
S2R with iterative data collection. An alternative way to overcome the problem of S2R gap
is learning distributions of dynamics parameters in an iterative manner. For example, Christiano
et al. [2016] learn inverse dynamics models from data gradually collected from a real robotic
system, while transferring trajectory planning policy from a simulator. Chebotar et al. [2019], Zhu
et al. [2018] transfer manipulation policies by iteratively collecting data on the real system and
updating a distribution of dynamics parameters for the simulator physics engine. Similar principles
work for the problem of humanoid balancing [Tan et al., 2016].
The common problem of these approaches is the necessity to execute untrained policies directly
on the robot, which may raise safety concerns. In contrast to all of the works presented above, we
are interested in a method that can i) transfer very low-level policies that directly control actuator
forces, ii) transfer to multiple real robots with different dynamics, iii) control inherently unstable
systems with dangerous consequences of failures, iv) avoid the need for data collection on the
real system with a possibly unstable policy, and v) avoid complex setups for system parameter
estimation.
Domain randomization. Domain randomization (DR) [Tobin et al., 2017] is a simple al-beit
promising domain adaptation technique that is well suited for S2R. It compensates for the
69
Object Description
| Title | Data scarcity in robotics: leveraging structural priors and representation learning |
| Author | Molchanov, Artem |
| Author email | a.molchanov86@gmail.com;molchano@usc.edu |
| Degree | Doctor of Philosophy |
| Document type | Dissertation |
| Degree program | Computer Science |
| School | Viterbi School of Engineering |
| Date defended/completed | 2020-05-11 |
| Date submitted | 2020-08-11 |
| Date approved | 2020-08-11 |
| Restricted until | 2020-08-11 |
| Date published | 2020-08-11 |
| Advisor (committee chair) | Sukhatme, Gaurav Suhas |
| Advisor (committee member) |
Ayanian, Nora Culbertson, Heather Gupta, Satyandra K. |
| Abstract | Recent advances in Artificial Intelligence have benefited significantly from access to large pools of data accompanied in many cases by labels, ground truth values, or perfect demonstrations. In robotics, however, such data are scarce or absent completely. Overcoming this issue is a major barrier to move robots from structured laboratory settings to the unstructured real world. In this dissertation, by leveraging structural priors and representation learning, we provide several solutions when data required to operate robotics systems is scarce or absent. ❧ In the first part of this dissertation we study sensory feedback scarcity. We show how to use high-dimensional alternative sensory modalities to extract data when primary sensory sources are absent. In a robot grasping setting, we address the problem of contact localization and solve it using multi-modal tactile feedback as the alternative source of information. We leverage multiple tactile modalities provided by electrodes and hydro-acoustic sensors to structure the problem as spatio-temporal inference. We employ the representational power of neural networks to acquire the complex mapping between tactile sensors and the contact locations. We also investigate scarce feedback due to the high cost of measurements. We study this problem in a challenging field robotics setting where multiple severely underactuated aquatic vehicles need to be coordinated. We show how to leverage collaboration among the vehicles and the spatio-temporal smoothness of the ocean currents as a prior to densify feedback about ocean currents in order to acquire better controllability. ❧ In the second part of this dissertation, we investigate scarcity of the data related to the desired task. We develop a method to efficiently leverage simulated dynamics priors to perform sim-to-real transfer of a control policy when no data about the target system is available. We investigate this problem in the scenario of sim-to-real transfer of low-level stabilizing quadrotor control policies. We demonstrate that we can learn robust policies in simulation and transfer them to the real system while acquiring no samples from the real quadrotor. Finally, we consider the general problem of learning a model with a very limited number of samples using meta-learned losses. We show how such losses can encode a prior structure about families of tasks to create well-behaved loss landscapes for efficient model optimization. We demonstrate the efficiency of our approach for learning policies and dynamics models in multiple robotics settings. |
| Keyword | robotics; machine learning; artificial intelligence |
| Language | English |
| Part of collection | University of Southern California dissertations and theses |
| Publisher (of the original version) | University of Southern California |
| Place of publication (of the original version) | Los Angeles, California |
| Publisher (of the digital version) | University of Southern California. Libraries |
| Provenance | Electronically uploaded by the author |
| Type | texts |
| Legacy record ID | usctheses-m |
| Contributing entity | University of Southern California |
| Rights | Molchanov, Artem |
| Physical access | The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given. |
| Repository name | University of Southern California Digital Library |
| Repository address | USC Digital Library, University of Southern California, University Park Campus MC 7002, 106 University Village, Los Angeles, California 90089-7002, USA |
| Repository email | cisadmin@lib.usc.edu |
| Filename | etd-MolchanovA-8923.pdf |
| Archival file | Volume13/etd-MolchanovA-8923.pdf |
Description
| Title | Page 84 |
| Full text | S2R with model parameter estimation. A substantial body of work considers closing the S2R gap by carefully estimating parameters of the real system to achieve a more realistic simulation. For example, Lowrey et al. [2018] transfer a non-prehensile manipulation policy for a system of three one-finger robot arms pushing a single cylinder. Tan et al. [2018] show transferability of agile locomotion gaits for quadruped robots. Antonova et al. [2017] learn a robust policy for rotating an object to a desired angle. While careful parameter estimation can provide a good estimate of the model, it often requires sophisticated setups [F¨orster, 2015]. In contrast, we transfer a learned policy to novel quadrotors for which we do not perform accurate model parameter estimation. S2R with iterative data collection. An alternative way to overcome the problem of S2R gap is learning distributions of dynamics parameters in an iterative manner. For example, Christiano et al. [2016] learn inverse dynamics models from data gradually collected from a real robotic system, while transferring trajectory planning policy from a simulator. Chebotar et al. [2019], Zhu et al. [2018] transfer manipulation policies by iteratively collecting data on the real system and updating a distribution of dynamics parameters for the simulator physics engine. Similar principles work for the problem of humanoid balancing [Tan et al., 2016]. The common problem of these approaches is the necessity to execute untrained policies directly on the robot, which may raise safety concerns. In contrast to all of the works presented above, we are interested in a method that can i) transfer very low-level policies that directly control actuator forces, ii) transfer to multiple real robots with different dynamics, iii) control inherently unstable systems with dangerous consequences of failures, iv) avoid the need for data collection on the real system with a possibly unstable policy, and v) avoid complex setups for system parameter estimation. Domain randomization. Domain randomization (DR) [Tobin et al., 2017] is a simple al-beit promising domain adaptation technique that is well suited for S2R. It compensates for the 69 |
