Probabilistic Movement Primitives

By Alexandros Paraschos, Christian Daniel, Jan Peters, Gerhard Neumann. 2013

728 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Benchmark for Skill Learning from Demonstration: Impact of User Experience, Task Complexity, and Start Configuration on Performance (2019, background): “Based on the choice of motion generation strategy, the majority of skill learning approaches can be further broadly categorized into four groups: time-invariant (stable) dynamical systems [2], [3], [13], time-dependent dynamical systems [7], [9], [10], time-dependent statistical methods [8], [14], [”
An Algorithmic Perspective on Imitation Learning (2018, methodology): “Probabilistic Movement Primitives (ProMPs) proposed by Paraschos et al. [2013] represent movement as a distribution over trajectories.”
Neural probabilistic motor primitives for humanoid control (2018, background): “These approaches can be seen as specific implementation choices for a certain notion of motor primitive, which emphasize the parameterization and learning of movement trajectories from repeated demonstrations (Paraschos et al., 2013; Meier & Schaal, 2016), rather than learning the actuation/stabiliz”
Learning position and orientation dynamics from demonstrations via contraction analysis (2018, background): “The idea of movement primitives is extended to probabilistic framework in Paraschos (2013).”
Learning Complex Neural Network Policies with Trajectory Optimization (2014, background): “Research into new, specialized policy classes is an active area that has provided substantial improvements on realworld systems (Ijspeert et al., 2003; Paraschos et al., 2013).”
A tutorial on task-parameterized movement learning and retrieval (2015, background): “The structure of the trajectory distribution defined in [71] requires multiple trajectory demonstrations to avoid overfitting, but the problem can be circumvented by employing factorization and variational inference techniques [76].”
Guiding Trajectory Optimization by Demonstrated Distributions (2017, methodology): “developed a framework called ProMP to model the distribution of the trajectory in parameter space [10].”
Recent Advances in Robot Learning from Demonstration (2020, background): “Stochastic policies sample a behavior from a learned distribution of behaviors during each execution (62, 100, 106).”
Exploiting Symmetries in Reinforcement Learning of Bimanual Robotic Tasks (2019, background): “Probabilistic Movement Primitives [4] (ProMPs) are further capable of capturing and reproducing the variance along time of a set of demonstrations.”
Geometrical Invariance and Smoothness Maximization for Task-Space Movement Generation (2016, methodology): “The method presented below does not attempt to use the human demonstration to infer a movement policy as in other recent approaches [48]–[51].”
Trajectory-Based Skill Learning Using Generalized Cylinders (2018, methodology): “Other approaches such as Probabilistic Movement Primitives (ProMP) (Paraschos et al., 2013) and Combined Learning from demonstration and Motion Planning (CLAMP) (Rana et al., 2017) approximate a stochastic control policy in the form of a Gaussian Process.”
MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies (2019, background): “Learning reusable representations that are transferable across multiple tasks has a long history in machine learning [1, 5, 30, 33, 43].”
A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms (2019, background): “ward probabilistic variant of DMPs that can produce distributions of trajectories [286].”
Towards Robust Skill Generalization: Unifying Learning from Demonstration and Motion Planning (2017, unknown): “A similar approach, probabilistic movement primitives (ProMPs) [14] directly fits a Gaussian distribution over demonstrations.”
Kernelized movement primitives (2017, unknown): “…t ∗ ∈ [0 , t D ] , we use τ ( t ∗ ) as the input for the prediction through KMP, and thus trajectories can be modulated as faster or slower (see also Ijspeert et al. (2013); Paraschos et al. (2013) for the modulations in time-scale, where the time modulation is called the phase transformation.).”
Learning Partially Contracting Dynamical Systems from Demonstrations (2017, background): “The idea of movement primitives is extended to probabilistic framework in [31].”