TY - GEN
T1 - Learning Task-Specific Dynamics to Improve Whole-Body Control
AU - Gams, Andrej
AU - Mason, Sean A.
AU - Ude, Ales
AU - Schaal, Stefan
AU - Rigbetti, Ludovic
N1 - Funding Information:
*This work was supported by the Slovenian Research Agency project BI-US/16-17-063, New York University, the Max-Planck Society and the European Unions Horizon 2020 research and innovation programme (grant agreement No 780684 and European Research Councils grant No 637935). 1Humanoid and Cognitive Robotics Lab, Dept. of Automatics, Bio-cybernetics and Robotics, Jozˇef Stefan Institute, Ljubljana, Slovenia. [email protected] 2Computational Learning and Motor Control Lab, University of Southern California, Los Angeles, California, USA. [email protected] 3Tandon School of Engineering, New York University, New York, USA. [email protected] 4Max Planck Institute for Intelligent Systems, Tuebingen, Germany.
Publisher Copyright:
© 2018 IEEE.
PY - 2018/7/2
Y1 - 2018/7/2
N2 - In task-based inverse dynamics control, reference accelerations used to follow a desired plan can be broken down into feedforward and feedback trajectories. The feedback term accounts for tracking errors that are caused from inaccurate dynamic models or external disturbances. On underactuated, free-floating robots, such as humanoids, good tracking accuracy often necessitates high feedback gains, which leads to undesirable stiff behaviors. The magnitude of these gains is anyways often strongly limited by the control bandwidth. In this paper, we show how to reduce the required contribution of the feedback controller by incorporating learned task-space reference accelerations. Thus, we i) improve the execution of the given specific task, and ii) offer the means to reduce feedback gains, providing for greater compliance of the system. In contrast to learning task-specific joint-torques, which might produce a similar effect but can lead to poor generalization, our approach directly learns the task-space dynamics of the center of mass of a humanoid robot. Simulated and real-world results on the lower part of the Sarcos Hermes humanoid robot demonstrate the applicability of the approach.
AB - In task-based inverse dynamics control, reference accelerations used to follow a desired plan can be broken down into feedforward and feedback trajectories. The feedback term accounts for tracking errors that are caused from inaccurate dynamic models or external disturbances. On underactuated, free-floating robots, such as humanoids, good tracking accuracy often necessitates high feedback gains, which leads to undesirable stiff behaviors. The magnitude of these gains is anyways often strongly limited by the control bandwidth. In this paper, we show how to reduce the required contribution of the feedback controller by incorporating learned task-space reference accelerations. Thus, we i) improve the execution of the given specific task, and ii) offer the means to reduce feedback gains, providing for greater compliance of the system. In contrast to learning task-specific joint-torques, which might produce a similar effect but can lead to poor generalization, our approach directly learns the task-space dynamics of the center of mass of a humanoid robot. Simulated and real-world results on the lower part of the Sarcos Hermes humanoid robot demonstrate the applicability of the approach.
UR - http://www.scopus.com/inward/record.url?scp=85062298203&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85062298203&partnerID=8YFLogxK
U2 - 10.1109/HUMANOIDS.2018.8624970
DO - 10.1109/HUMANOIDS.2018.8624970
M3 - Conference contribution
AN - SCOPUS:85062298203
T3 - IEEE-RAS International Conference on Humanoid Robots
SP - 658
EP - 663
BT - 2018 IEEE-RAS 18th International Conference on Humanoid Robots, Humanoids 2018
PB - IEEE Computer Society
T2 - 18th IEEE-RAS International Conference on Humanoid Robots, Humanoids 2018
Y2 - 6 November 2018 through 9 November 2018
ER -