Award Abstract # 1544857
CPS: Synergy: Learning to Walk - Optimal Gait Synthesis and Online Learning for Terrain-Aware Legged Locomotion

NSF Org: CNS
Division Of Computer and Network Systems
Recipient: GEORGIA TECH RESEARCH CORP
Initial Amendment Date: September 17, 2015
Latest Amendment Date: September 14, 2017
Award Number: 1544857
Award Instrument: Continuing Grant
Program Manager: David Corman
CNS
 Division Of Computer and Network Systems
CSE
 Directorate for Computer and Information Science and Engineering
Start Date: October 1, 2015
End Date: September 30, 2019 (Estimated)
Total Intended Award Amount: $799,982.00
Total Awarded Amount to Date: $799,982.00
Funds Obligated to Date: FY 2015 = $556,775.00
FY 2016 = $91,006.00

FY 2017 = $152,201.00
History of Investigator:
  • Patricio Vela (Principal Investigator)
    pvela@ece.gatech.edu
  • Erik Verriest (Co-Principal Investigator)
  • Daniel Goldman (Co-Principal Investigator)
  • Aaron Ames (Co-Principal Investigator)
Recipient Sponsored Research Office: Georgia Tech Research Corporation
926 DALNEY ST NW
ATLANTA
GA  US  30318-6395
(404)894-4819
Sponsor Congressional District: 05
Primary Place of Performance: Georgia Tech Research Corporation
85 5th ST NW
Atlanta
GA  US  30332-0250
Primary Place of Performance
Congressional District:
05
Unique Entity Identifier (UEI): EMW9FC8J3HN4
Parent UEI: EMW9FC8J3HN4
NSF Program(s): Special Projects - CNS,
IIS Special Projects,
CPS-Cyber-Physical Systems
Primary Program Source: 01001516DB NSF RESEARCH & RELATED ACTIVIT
01001617DB NSF RESEARCH & RELATED ACTIVIT

01001718DB NSF RESEARCH & RELATED ACTIVIT
Program Reference Code(s): 7918, 8235
Program Element Code(s): 171400, 748400, 791800
Award Agency Code: 4900
Fund Agency Code: 4900
Assistance Listing Number(s): 47.070

ABSTRACT

Legged robots have captured the imagination of society at large, through
entertainment and through the dissemination of research findings. Yet,
today's reality of what (bipedal) legged robots can do falls short of
society's vision. A big part of the reason is that legged robots are
viewed as surrogates for humans, able to go wherever humans can as aids
or as assistants where it might also be too dangerous or risky. It is
in the expectation of robustness and walking facility that today's
research hits its limits, especially when the terrain has granular
properties. Impeding progress is the lack of a holistic approach to the
cyber-physical modeling and control of legged robots. The vision of
this work is to unite experts in granular mechanics, optimal control,
and learning theory in order to define a methodology for advancing
cyber-physical systems (CPS) involving a tight coupling of the physical with
the cyber through dynamic interactions that must be learned online. The
proposed work will advance the science of cyber-physical systems by more
explicitly tying sensing, perception, and computing to the optimization
and control of physical systems whose properties are variable and
uncertain. Achieving reliable, adaptive legged locomotion over terrain
with arbitrary granular properties would transform several application
domain areas of robotics; e.g., disaster response, agricultural and
industrial robotics, and planetary robotics. More broadly, the same
tools would apply to related CPS with regards to terrain aware
exoskeleton and rehabilitation prosthetics for persons with missing,
non-functional, or injured legs, as well as to energy networks with
time-varying, nonlinear dynamics models.


The CPS platform to be studied is that of a bipedal robot locomoting
over granular ground material with uncertain physical properties (sand,
gravel, dirt, etc.). The proposed work seeks to overcome current
impediments to reliable legged locomotion over uncertain terrain type,
which fundamentally relies on the controlled interaction of the robot's
feet with the physical environment. The research goal is to improve the
perception and control of legged locomotion over granular media for the
express purpose of achieving robust, adaptive, terrain-aware locomotion.
It revolves around the hypothesis that simple models with decent
predictive performance and low computational overhead are sufficient for
the optimal control formulations as the compute-constrained adaptive
subsystem will both learn and classify the peculiarities of the terrain
online. The main research objectives will involve: [1] a validated
co-simulation platform for legged robot movement over granular media;
[2] terrain-dependent, stable gait generation and gait transition
strategies via optimal control; [3] online, compute-constrained learning
of granular interactions for adaptation and terrain classification; and
[4] validated contributions using experimental testbeds involving
variable and unknown (to the robot) granular media. Given the high
value of the robotic platforms and the research with regards to outreach
and participation, they will be used as outreach tools and to create new
educational modules for promotion of STEM fields. Further, the
multi-disciplinary nature of the work will be highlighted in order to
emphasize its importance.

PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH

Note:  When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

(Showing: 1 - 10 of 18)
Aguilar, Jeffrey and Goldman, Daniel I "Robophysical study of jumping dynamics on granular media" Nature Physics , v.12 , 2015 , p.278 www.nature.com/nphys/journal/v12/n3/pdf/nphys3568.pdf
A. H. Chang and C. M. Hubicki and J. J. Aguilar and D. I. Goldman and A. D. Ames and P. A. Vela "Learning to jump in granular media: Unifying optimal control synthesis with Gaussian process-based regression" IEEE International Conference on Robotics and Automation , 2017 , p.2154 https://doi.org/10.1109/ICRA.2017.7989248
Erik I. Verriest and Vishal Murali "Graceful Transitions between Limit Cycles of a Parameterized System" American Control Conference , 2018 , p.6075-6080 10.23919/ACC.2018.8431126
Hubicki C.M., Aguilar J.J., Goldman D.I., Ames A.D. "Tractable Terrain-aware Motion Planning on Granular Media: An Impulsive Jumping Study." IEEE International Conference on Intelligent Robots and Systems , 2016 , p.3887 https://doi.org/10.1109/IROS.2016.7759572
K. Sakurama and E.I. Verriest and M. Egerstedt "Scalable Stability and Time-Scale Separation ofNetworked, Cascaded Systems" IEEE Transactions on Control of Network Systems , v.5 , 2018 10.1109/TCNS.2016.2609146
L. D. Fairfax and P. A. Vela "A Concurrent Learning Approach to Monocular, Vision-Based Regulation of Leader/Follower Systems" Annual American Control Conference , 2018 , p.3502-3507 10.23919/ACC.2018.8430812
Qian, Feifei and Zhang, Tingnan and Korff, Wyatt and Umbanhowar, Paul B and Full, Robert J and Goldman, Daniel I "Principles of appendage design in robots and animals determining terradynamic performance on flowable ground" Bioinspiration \& biomimetics , v.10 , 2015 , p.056014 http://dx.doi.org/10.1088/1748-3190/10/5/056014
Reher, Jenna and Ames, Aaron D. "Dynamic Walking: Toward Agile and Efficient Bipedal Robots" Annual Review of Control, Robotics, and Autonomous Systems , v.4 , 2021 https://doi.org/10.1146/annurev-control-071020-045021 Citation Details
V. Azhmyakov and E.I. Verriest and L.A. Guzman Trujilloand S. Pickl "On the Optimal Control of Multidimensional DynamicSystems Evolving with State Suprema" IEEE Conference on Decision and Control , 2018 10.1109/CDC.2018.8618934
V. Azhmyakov and J.P. Fernandez-Gutierrez and E.I. Verriest and S.W. Pickl "A Separation Based Optimization Approach to DynamicMaximal Covering Location Problems with Switched Structure" Journal of Industrial and Management Optimization , 2019 10.3934/jimo.2019128
V. Azimi and P. A. Vela "Performance Reference Adaptive Control: A Joint Quadratic Programming and Adaptive Control Framework" Annual American Control Conference , 2018 , p.1827-1834 10.23919/ACC.2018.8431150
(Showing: 1 - 10 of 18)

PROJECT OUTCOMES REPORT

Disclaimer

This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.

The main goal of the project was to advance our understanding of dynamic walking of bipedal robots over more diverse terrain profiles than hard ground. When traversing yielding ground, the ability to perceive, learn, and respond to variable terrain reaction forces is important for preserving the stability of walking. This problem relates to cyber-physical systems because sensing and control must work together, with learning and adaptation being the link, to realize successful walking over uncertain terrain while being computationally tractable. To advance on this front, the project was envisioned to provide four contributions. The first was to obtain faithful models of foot-substrate interaction for use by control synthesis solvers, which were to later use learning-informed interaction models. Related to this, the second was to derive computationally efficient, terrain-dependent optimal control strategies based on low-fidelity models, with the ability to transition across different gait classes. On these two fronts, we were able to successfully demonstrate hopping and walking over granular media based on the granular models derived during the project, both in simulation and on experimental testbeds. Furthermore, we demonstrated in simulation the ability to optimally transition between different gaits and between different known terrain profiles. The transition method has favorable properties over other existing gait transition methods in the it attempts to preserve important underlying properties of the gait during the transition phase.

The third objective was to dispense with the terrain model and learning the foot-substrate interaction forces through experience. If possible, the terrain was to be characterized or classified for rapid adaptation to known terrain classes. We demonstrated the ability to rapidly learn in a few-shot manner the unknown interaction forces and to optimize actuation to achieve a specified and achievable control task (i.e., a target hopping height). In addition to learning the interaction forces, these same models permitted classification of novel experiences to the closest matching terrain.  When the terrain was previously experienced, hopping tasks could be achieved immediately.  When it was novel, the nearest classification sped up learning and required fewer iterations to meet the task objective. Overall, the project objectives were successfully achieved and confirmed on increasingly complex legged platforms, including a hopper robot, a small bipedal robots, and two larger, human height bipedal robots.

Beyond the research efforts, we also exposed local Atlanta and Pasadena youth to robotics, engaged several undergraduates in engineering research, and translated select problems to the classroom.  Activities targeting general audiences and videos posted online also communicated the value of our findings to self-learners of all ages.

 


Last Modified: 02/16/2020
Modified by: Patricio A Vela

Please report errors in award information by writing to: awardsearch@nsf.gov.

Print this page

Back to Top of page