Learning a Multi-Task Model for Control of a Compliant Robot

Many strategies for controlling a robot require a detailed model of the robot’s kinematics or dynamics. However, such models can be dependent on unknown context variables. One such context variable is the mass of an object lifted by the robot, since for compliant robots this mass changes the model in a non-linear way. Models for specific contexts can be learned using Gaussian Processes (GPs), but have to be relearned for each novel context. In this paper multi-task GPs are used to learn a model that allows generalization between different contexts. Consequently, only a few new data points have to be generated when a new context is encountered. We employ learned task features to efficiently learn the multi-task model. The approach is evaluated on an object-lifting task with a continuum robot arm under several settings, and compared to standard GPs trained for each object separately. The learned model is shown to be usable for control. We show that multitask GPs using learned feature representations for the tasks outperform standard GPs when few data points are available.

[1]  J. Y. S. Luh,et al.  An anatomy of industrial robots and their controls , 1983 .

[2]  Edwin V. Bonilla,et al.  Multi-task Gaussian Process Prediction , 2007, NIPS.

[3]  Anton Schwaighofer,et al.  Learning Gaussian processes from multiple tasks , 2005, ICML.

[4]  O. Sawodny,et al.  Forward kinematics of a compliant pneumatically actuated redundant manipulator , 2012, 2012 7th IEEE Conference on Industrial Electronics and Applications (ICIEA).

[5]  Sethu Vijayakumar,et al.  Load estimation and control using learned dynamics models , 2007, 2007 IEEE/RSJ International Conference on Intelligent Robots and Systems.

[6]  David J. Fleet,et al.  Multifactor Gaussian process models for style-content separation , 2007, ICML '07.

[7]  Sebastian Thrun,et al.  Is Learning The n-th Thing Any Easier Than Learning The First? , 1995, NIPS.

[8]  N. Higham.  COMPUTING A NEAREST SYMMETRIC POSITIVE SEMIDEFINITE MATRIX , 1988 .

[9]  Farbod Fahimi,et al.  Autonomous Robots: Modeling, Path Planning, and Control , 2008 .

[10]  Carl E. Rasmussen,et al.  Policy search for learning robot control using sparse data , 2014, 2014 IEEE International Conference on Robotics and Automation (ICRA).

[11]  In Ha,et al.  An approach to nonlinear feedback control with applications to robotics , 1984, The 22nd IEEE Conference on Decision and Control.

[12]  Fabio Tozeto Ramos,et al.  Multi-task learning of system dynamics with maximum information gain , 2011, 2011 IEEE International Conference on Robotics and Automation.

[13]  Qiang Yang,et al.  A Survey on Transfer Learning , 2010, IEEE Transactions on Knowledge and Data Engineering.

[14]  Pushparaj Mani Pathak,et al.  Modelling of multisection bionic manipulator: Application to RobotinoXT , 2011, 2011 IEEE International Conference on Robotics and Biomimetics.

[15]  Jochen J. Steil,et al.  Efficient Exploratory Learning of Inverse Kinematics on a Bionic Elephant Trunk , 2014, IEEE Transactions on Neural Networks and Learning Systems.

[16]  Sethu Vijayakumar,et al.  Multi-task Gaussian Process Learning of Robot Inverse Dynamics , 2008, NIPS.

[17]  Jan Peters,et al.  Model Learning with Local Gaussian Process Regression , 2009, Adv. Robotics.

[18]  Edwin V. Bonilla,et al.  Kernel Multi-task Learning using Task-specific Features , 2007, AISTATS.

[19]  Jan Peters,et al.  Model learning for robot control: a survey , 2011, Cognitive Processing.

[20]  Neil D. Lawrence,et al.  Gaussian Process Latent Variable Models for Visualisation of High Dimensional Data , 2003, NIPS.

[21]  D M Wolpert,et al.  Multiple paired forward and inverse models for motor control , 1998, Neural Networks.