Optimizing robot striking movement primitives with Iterative Learning Control

Koc, Okan; Maeda, Guilherme; Neumann, Gerhard; Peters, Jan

doi:10.1109/humanoids.2015.7363535

Cited by 5 publications

(8 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We have compared bayesILC to two other ILC methods: batch ILC (32) and ILC with PD feedback (with constant p, d values). PD type ILC with constant p or d values is often too simplistic, and did not yield any improvement in our setup, even after tuning the p, d values.…”

Section: B Real Robot Table Tennismentioning

confidence: 99%

“…However, these approaches are heavily structured for the problem at hand, introducing and tuning additional domain parameters. In [32], we instead proposed to use rhythmic movement primitives that allow for a limit cycle attractor, which is desirable if we want to maintain the striking motion through goal state. After the striking is completed the DMP can be used to return back to initial state or it can be terminated by setting the forcing terms to zero.…”

Section: Appendix B Movement Generation For Table Tennismentioning

confidence: 99%

“…CI adds the current iteration k's feedback errors to the feedforward compensations for the next iteration k + 1. As in our preliminary experiments with the Barrett WAM[32], we have applied CI in addition to stabilize a downsampled version of batch model-based ILC.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Optimizing the Execution of Dynamic Robot Movements With Learning Control

2019

Self Cite

View full text Add to dashboard Cite

High-speed robotics typically involves fast dynamic trajectories with large accelerations. Kinematic optimization using compact representations can lead to an efficient online computation of these dynamic movements, however successful execution requires accurate models or aggressive tracking with high-gain feedback. Learning to track such references in a safe and reliable way, whenever accurate models are not available, is an open problem. Stability issues surrounding the learning performance, in the iteration domain, can prevent the successful implementation of model-based learning approaches. To this end, in this paper we propose a new adaptive and cautious iterative learning control (ILC) algorithm where the stability of the control updates is analyzed probabilistically: the covariance estimates of the adapted local linear models are used to increase the probability of update monotonicity, exercising caution during learning. The resulting learning controller can be implemented efficiently using a recursive approach. We evaluate it extensively in simulations as well as in our robot table tennis setup for tracking dynamic hitting movements. Testing with two seven degree of freedom anthropomorphic robot arms, we show improved and more stable tracking performance over high-gain proportional and derivative (PD) control, model-free ILC (simple PD feedback type) and model-based ILC without cautious adaptation.

show abstract

Section: B Real Robot Table Tennismentioning

confidence: 99%

Section: Appendix B Movement Generation For Table Tennismentioning

confidence: 99%

See 1 more Smart Citation

Optimizing the Execution of Dynamic Robot Movements With Learning Control

2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…The Frobenius norm of the trajectory deviations, J k , is plotted over the iterations k. Results are averaged over ten experiments, where for each experiment, trajectories, nominal models and actual models are drawn randomly from Gaussian Processes. The performance of the batch pseudoinverse ILC (32) is shown in the red line. Numerical stability issues prevent it from stabilizing at steady state error, whereas recursive ILC (blue line) converges stably.…”

Section: Evaluations and Experimentsmentioning

confidence: 99%

Optimizing Execution of Dynamic Goal-Directed Robot Movements with Learning Control

Koc,

Maeda,

Peters

2018

Preprint

Self Cite

View full text Add to dashboard Cite

High-speed robotics typically involves fast dynamic trajectories with large accelerations. Kinematic optimization using compact representations can lead to an efficient online computation of these dynamic movements, however successful execution requires accurate models or aggressive tracking with high-gain feedback. Learning to track such references in a safe and reliable way, whenever accurate models are not available, is an open problem. Stability issues surrounding the learning performance, in the iteration domain, can prevent the successful implementation of model-based learning approaches. To this end, we propose a new adaptive and cautious Iterative Learning Control (ILC) algorithm where the stability of the control updates is analyzed probabilistically: the covariance estimates of the adapted local linear models are used to increase the probability of update monotonicity, exercising caution during learning. The resulting learning controller can be implemented efficiently using a recursive approach. We evaluate it extensively in simulations as well as in our robot table tennis setup for tracking dynamic hitting movements. Testing with two seven degree of freedom anthropomorphic robot arms, we show improved and more stable tracking performance over high-gain PD-control, model-free ILC (simple PD feedback type) and model-based ILC without cautious adaptation.

show abstract

“…Robot table tennis has, since the nineties, captivated the attention of the robot control and learning communities as a challenging and dynamic task, and research in it has been ongoing ever since. After the pioneering work of Anderson's analytical player [1], there have been various approaches focusing on certain parts of the game, such as simplifications in trajectory generation using a virtual hitting plane [2], [3] or learning striking trajectories from demonstrations [4]. Learning approaches to generate better strikes with Reinforcement Learning (RL) include [5], [6].…”

Section: Introductionmentioning

confidence: 99%

Learning to Serve: An Experimental Study for a New Learning From Demonstrations Framework

Koc

Peters

2019

IEEE Robot. Autom. Lett.

Self Cite

View full text Add to dashboard Cite

Learning from demonstrations is an easy and intuitive way to show examples of successful behavior to a robot. However, the fact that humans optimize or take advantage of their body and not of the robot, usually called the embodiment problem in robotics, often prevents industrial robots from executing the task in a straightforward way. The shown movements often do not or cannot utilize the degrees of freedom of the robot efficiently, and moreover can suffer from excessive execution errors. In this letter, we explore a variety of solutions that address these shortcomings. In particular, we learn sparse movement primitive parameters from several demonstrations of a successful table tennis serve. The number of parameters learned using our procedure is independent of the degrees of freedom of the robot. Moreover, they can be ranked according to their importance in the regression task. Learning few parameters, which are ranked, is a desirable feature to combat the curse of dimensionality in reinforcement learning. Real robot experiments on the Barrett WAM for a table tennis serve using the learned movement primitives show that the representation can capture successfully the style of the movement with few parameters.

show abstract

Optimizing robot striking movement primitives with Iterative Learning Control

Cited by 5 publications

References 28 publications

Optimizing the Execution of Dynamic Robot Movements With Learning Control

Optimizing the Execution of Dynamic Robot Movements With Learning Control

Optimizing Execution of Dynamic Goal-Directed Robot Movements with Learning Control

Learning to Serve: An Experimental Study for a New Learning From Demonstrations Framework

Contact Info

Product

Resources

About