Bayesian uncertainty modeling for programming by demonstration

Umlauft, Jonas; Fanger, Yunis; Hirche, Sandra

doi:10.1109/icra.2017.7989759

Cited by 13 publications

(13 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The work in [36] proposes a feedback linearizing control law, which adapts online the weights of a neural network model. It shows boundedness of the adaptation law and the resulting controller but cannot quantify the ultimate bound because neural networks -in comparison to GPs -do not inherently provide a measure for the fidelity of the model [37]. This becomes important, if the controller is applied in safety critical domains, where the tracking error must be quantified to avoid failure or damage to the system.…”

Section: A Related Workmentioning

confidence: 99%

Feedback Linearization Based on Gaussian Processes With Event-Triggered Online Learning

Umlauft

Hirche

2020

IEEE Trans. Automat. Contr.

Self Cite

View full text Add to dashboard Cite

Combining control engineering with nonparametric modeling techniques from machine learning allows to control systems without analytic description using data-driven models. Most existing approaches separate learning, i.e. the system identification based on a fixed dataset, and control, i.e. the execution of the model-based control law. This separation makes the performance highly sensitive to the initial selection of training data and possibly requires very large datasets. This article proposes a learning feedback linearizing control law using online closed-loop identification. The employed Gaussian process model updates its training data only if the model uncertainty becomes too large. This event-triggered online learning ensures high data efficiency and thereby reduces the computational complexity, which is a major barrier for using Gaussian processes under realtime constraints. We propose safe forgetting strategies of data points to adhere to budget constraint and to further increase data-efficiency. We show asymptotic stability for the tracking error under the proposed event-triggering law and illustrate the effective identification and control in simulation.

show abstract

Section: A Related Workmentioning

confidence: 99%

Feedback Linearization Based on Gaussian Processes With Event-Triggered Online Learning

Umlauft

Hirche

2020

IEEE Trans. Automat. Contr.

Self Cite

View full text Add to dashboard Cite

show abstract

“…they do not encode input-dependent covariance. Works such as [10], [17], which propose formulations for regressing both uncertainty and correlation, may provide the possibility to simultaneously consider both aspects into our approach, which can potentially allow us to study the complementarity of the two Fig. 7: Stiffness and damping gains along x 2 , estimated for three points at increasing distances from the training data.…”

Section: Discussion and Future Workmentioning

confidence: 99%

“…Recently, this notion and its implications in practical scenarios have been debated to some extent as different state-of-the-art probabilistic techniques provide complementary notions of variance. In [10], Umlauft et al discuss the differences between variance being interpreted as uncertainty and variability. The topic is also covered in [11], where the different notions are exploited in scenarios that require the combination of different controllers, and in [12], in the context of robot dynamics with multiple additive noise sources.…”

Section: Related Workmentioning

confidence: 99%

An Uncertainty-Aware Minimal Intervention Control Strategy Learned from Demonstrations

Silvério

Huang

Rozo

2018

2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

Motivated by the desire to have robots physically present in human environments, in recent years we have witnessed an emergence of different approaches for learning active compliance. Some of the most compelling solutions exploit a minimal intervention control principle, correcting deviations from a goal only when necessary, and among those who follow this concept, several probabilistic techniques have stood out from the rest. However, these approaches are prone to requiring several task demonstrations for proper gain estimation and to generating unpredictable robot motions in the face of uncertainty. Here we present a Programming by Demonstration approach for uncertainty-aware impedance regulation, aimed at making the robot compliant-and safe to interact with-when the uncertainty about its predicted actions is high. Moreover, we propose a data-efficient strategy, based on the energy observed during demonstrations, to achieve minimal intervention control, when the uncertainty is low. The approach is validated in an experimental scenario, where a human collaboratively moves an object with a 7-DoF torque-controlled robot.

show abstract

“…However, by benefiting from the KMP predictions, it unifies the best Types of prediction Variability Uncertainty Correlations GMM/GMR [3] -GPR [8] --HGP [15], [16] -MDN [17] -Our approach of the two approaches. Umlauft et al [18] propose a related formulation where, using Wishart processes, they build full covariance matrices with uncertainty. However their solution requires a very high number of parameters, whose estimation relies heavily on optimization, and their control gains are set heuristically.…”

Section: Related Workmentioning

confidence: 99%

Uncertainty-Aware Imitation Learning using Kernelized Movement Primitives

Silvério

Huang

Abu‐Dakka

et al. 2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

During the past few years, probabilistic approaches to imitation learning have earned a relevant place in the robotics literature. One of their most prominent features is that, in addition to extracting a mean trajectory from task demonstrations, they provide a variance estimation. The intuitive meaning of this variance, however, changes across different techniques, indicating either variability or uncertainty. In this paper we leverage kernelized movement primitives (KMP) to provide a new perspective on imitation learning by predicting variability, correlations and uncertainty using a single model. This rich set of information is used in combination with the fusion of optimal controllers to learn robot actions from data, with two main advantages: i) robots become safe when uncertain about their actions and ii) they are able to leverage partial demonstrations, given as elementary sub-tasks, to optimally perform a higher level, more complex task. We showcase our approach in a painting task, where a human user and a KUKA robot collaborate to paint a wooden board. The task is divided into two sub-tasks and we show that the robot becomes compliant (hence safe) outside the training regions and executes the two sub-tasks with optimal gains otherwise.

show abstract

Bayesian uncertainty modeling for programming by demonstration

Cited by 13 publications

References 10 publications

Feedback Linearization Based on Gaussian Processes With Event-Triggered Online Learning

Feedback Linearization Based on Gaussian Processes With Event-Triggered Online Learning

An Uncertainty-Aware Minimal Intervention Control Strategy Learned from Demonstrations

Uncertainty-Aware Imitation Learning using Kernelized Movement Primitives

Contact Info

Product

Resources

About