Learning of parametric coupling terms for robot-environment interaction

Gams, Andrej; Deniša, Miha; Ude, Aleš

doi:10.1109/humanoids.2015.7363559

Cited by 16 publications

(20 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recently, Gams el al. expanded their work in [12], by generating a database of coupling terms and generalizing to multiple scenarios. All of the above approaches take an iterative approach towards learning the parameters of their coupling term model, but suffer from a lack of generalizability to unseen settings.…”

Section: A Coupling Termsmentioning

confidence: 99%

Learning feedback terms for reactive planning and control

Rai¹,

Sutanto²,

Schaal³

et al. 2017

2017 IEEE International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

With the advancement of robotics, machine learning, and machine perception, increasingly more robots will enter human environments to assist with daily tasks. However, dynamically-changing human environments requires reactive motion plans. Reactivity can be accomplished through replanning, e.g. model-predictive control, or through a reactive feedback policy that modifies on-going behavior in response to sensory events. In this paper, we investigate how to use machine learning to add reactivity to a previously learned nominal skilled behavior. We approach this by learning a reactive modification term for movement plans represented by nonlinear differential equations. In particular, we use dynamic movement primitives (DMPs) to represent a skill and a neural network to learn a reactive policy from human demonstrations. We use the well explored domain of obstacle avoidance for robot manipulation as a test bed. Our approach demonstrates how a neural network can be combined with physical insights to ensure robust behavior across different obstacle settings and movement durations. Evaluations on an anthropomorphic robotic system demonstrate the effectiveness of our work. * both authors contributed equally to this work

show abstract

Section: A Coupling Termsmentioning

confidence: 99%

Learning feedback terms for reactive planning and control

Rai¹,

Sutanto²,

Schaal³

et al. 2017

2017 IEEE International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

show abstract

“…For example, Matsubara et al [19] proposed an algorithm for the generation of new control policies from existing knowledge, thereby achieving an extended scalability of DMPs, while mixture of motor primitives was used for generation of table tennis swings in [20]. On the other hand, generalization of DMPs was combined with model predictive control by Krug and Dimitrov [21] or applied to DMP coupling terms [22], which were learned and later added to a demonstrated trajectory to generate new joint space trajectories. Furthermore, Stulp et al [23] proposed learning a function approximator with one regression in the full space of phase and tasks parameters, bypassing the need for two consecutive regressions.…”

Section: Generalization Of Position and Orientation Trajectoriesmentioning

confidence: 99%

Generalization of orientation trajectories and force-torque profiles for robotic assembly

Kramberger

Gams

Nemec

et al. 2017

Robotics and Autonomous Systems

Self Cite

View full text Add to dashboard Cite

A typical robot assembly operation involves contacts with the parts of the product to be assembled and consequently requires the knowledge of not only position and orientation trajectories but also the accompanying force-torque profiles for successful performance. To learn the execution of assembly operations even when the geometry of the product varies across task executions, the robot needs to be able to adapt its motion based on a parametric description of the current task condition, which is usually provided by geometrical properties of the parts involved in the assembly. In our previous work we showed how positional control policies can be generalized to different task conditions. In this paper we propose a complete methodology to generalize also the orientational trajectories and the accompanying force-torque profiles to compute the necessary control policy for a given condition of the assembly task. Our method is based on statistical generalization of successfully recorded executions at different task conditions, which are acquired by kinesthetic guiding. The parameters that describe the varying task conditions define queries into the recorded training data. To improve the execution of the skill after generalization, we combine the proposed approach with an adaptation method, thus enabling the refinement of the generalized assembly operation.

show abstract

“…Generalization using GPR was applied over combined joint position torque trajectories in the framework of compliant movement primitives [12], extending the DMP framework beyond the kinematic trajectory properties. Generalization was also applied to DMP coupling terms [13], which were learned and later added to a demonstrated trajectory to generate new joint-space trajectories.…”

Section: Related Workmentioning

confidence: 99%

Generalization of orientational motion in unit quaternion space

Kramberger

Gams

Nemec

et al. 2016

2016 IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids)

Self Cite

View full text Add to dashboard Cite

A requirement for the humanoid robot's operation in natural environments is that a humanoid robot can effectively adapt to new configurations of the external world. In this paper we address the problem of adaptation, where new robot movements are generated based on data accumulated in related but different situations. Our solution to this challenge is to apply statistical learning, which provides a method to generate robot responses in new situations. Building on our previous work on learning motor primitives [1], we propose a new methodology for task-specific generalization of orientation trajectories, which we encode as Cartesian space Dynamic Movement Primitives. Example trajectories are generalized by applying Locally Weighted Regression in unit quaternion space, using the parameters describing the task as query points into the trajectory database. We show on real-world and simulated tasks that the proposed methodology can be used for statistical learning of orientation trajectories.

show abstract

Learning of parametric coupling terms for robot-environment interaction

Cited by 16 publications

References 18 publications

Learning feedback terms for reactive planning and control

Learning feedback terms for reactive planning and control

Generalization of orientation trajectories and force-torque profiles for robotic assembly

Generalization of orientational motion in unit quaternion space

Contact Info

Product

Resources

About