• Corpus ID: 228376399

Structured Policy Representation: Imposing Stability in arbitrarily conditioned dynamic systems

@article{Urain2020StructuredPR,
  title={Structured Policy Representation: Imposing Stability in arbitrarily conditioned dynamic systems},
  author={Julen Urain and Davide Tateo and Tianyu Ren and Jan Peters},
  journal={ArXiv},
  year={2020},
  volume={abs/2012.06224}
}
We present a new family of deep neural network-based dynamic systems. The presented dynamics are globally stable and can be conditioned with an arbitrary context state. We show how these dynamics can be used as structured robot policies. Global stability is one of the most important and straightforward inductive biases as it allows us to impose reasonable behaviors outside the region of the demonstrations. 

Figures from this paper

Towards A Unified Policy Abstraction Theory and Representation Learning Approach in Markov Decision Processes

A proposed policy abstraction theory, containing three types of policy abstraction associated to policy features at different levels, and a policy representation learning approach based on deep metric learning, which indicates that there is no a universally optimal abstraction for all downstream learning problems.

References

SHOWING 1-10 OF 15 REFERENCES

ImitationFlow: Learning Deep Stable Stochastic Dynamic Systems by Normalizing Flows

The Lyapunov stability for a class of Stochastic Differential Equations is proved and a learning algorithm is proposed to learn them from a set of demonstrated trajectories.

Learning Stable Nonlinear Dynamical Systems With Gaussian Mixture Models

A learning method is proposed, which is called Stable Estimator of Dynamical Systems (SEDS), to learn the parameters of the DS to ensure that all motions closely follow the demonstrations while ultimately reaching and stopping at the target.

Learning Partially Contracting Dynamical Systems from Demonstrations

The CDSP algorithm is shown to be capable of learning and reproducing point-to-point motions directly from real-world demonstrations using a Baxter robot and also compared with two state-of-the-art motion generation algorithms.

Euclideanizing Flows: Diffeomorphic Reduction for Learning Stable Dynamical Systems

This work presents an approach to learn such motions from a limited number of human demonstrations by exploiting the regularity properties of human motions e.g. stability, smoothness, and boundedness by exploiting a composition of simple parameterized diffeomorphisms.

Learning Contracting Vector Fields For Stable Imitation Learning

A rich family of smooth vector fields induced by certain classes of matrix-valued kernels is constructed, whose equilibria are placed exactly at a desired set of locations and whose local contraction and curvature properties at various points can be explicitly controlled using convex optimization.

Learning by Demonstration

  • S. Schaal
  • Education, Computer Science
    Encyclopedia of Machine Learning and Data Mining
  • 1996
In an implementation of pole balancing on a complex anthropomorphic robot arm, it is demonstrated that, when facing the complexities of real signal processing, model-based reinforcement learning offers the most robustness for LQR problems.