Learning Attractor Landscapes for Learning Motor Primitives

By Auke Jan Ijspeert, Jun Nakanishi, Stefan Schaal. 2002

747 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.

In Reinforcement Learning

Citations in the corpus

Learning Dynamic Robot-to-Human Object Handover from Human Feedback (2016, background): “To represent robot skills, policy search typically makes use of parametrized controllers, such as dynamic movement primitives [16] or interaction primitives [2].”
Accelerating Online Reinforcement Learning with Offline Datasets (2020, background): “sidered the special case of learning from demonstration data. One class of algorithms initializes the policy via behavioral cloning from demonstrations, and then ﬁne-tunes with reinforcement learning [33, 15, 40, 18, 35, 10, 50]. Most such methods use on-policy ﬁnetuning, which is less sample-efﬁcie”
Reinforcement learning in robotics: A survey (2013, background): “Dynamical system motor primitives (Ijspeert et al., 2003; Schaal et al., 2007) have become a popular representation for reinforcement learning of discrete movements.”
Dexterous Manipulation with Deep Reinforcement Learning: Efficient, General, and Low-Cost (2018, methodology): “community [17], [16], [18], [19], [1], [20], [21] but these methods have not been applied to real world manipulation with multi-fingered hands.”
An Algorithmic Perspective on Imitation Learning (2018, background): “Also, when transferring motions captured from a human expert to a humanoid robot the control inputs to achieve the desired motion in the learner’s embodiment cannot be observed [Ijspeert et al., 2002b, Grimes et al., 2006b, Grimes and Rao, 2009].”
Learning contact-rich manipulation skills with guided policy search (2015, background): “controller representation is related to other trajectory-centric representations in policy search, such as dynamic movement primitives (DMPs) [10].”
Deep predictive policy training using reinforcement learning (2017, background): “These approaches, known as dynamic movement primitives (DMP), have found applications in robot learning by demonstration [25] and also reinforcement learning [3], [26].”
Learning dexterous manipulation for a soft robotic hand from human demonstrations (2016, methodology): “A widely used approach is to learn the parameters of a dynamic motor primitive [10] (DMP) with relative entropy policy search [11] or PI2 [12].”
Active Reward Learning (2014, methodology): “A popular use case for PS methods is to find good parameters of trajectory generators such as Dynamic Movement Primitives (DMPs) [17].”
Meta Reinforcement Learning with Latent Variable Gaussian Processes (2018, background): “Much of the literature on meta and transfer learning in RL has focused on multi-task learning, i.e., cases where the system/robot is the same, but the task changes [16, 29, 3, 5, 20, 21, 24, 8, 12].”
Learning and Transfer of Modulated Locomotor Controllers (2016, background): “Motor primitives have resonated in robotics, especially as Dynamic Movement Primitives [11], which are low-dimensionality attractor systems that can simplify the learning of robot movements, and hierarchical robot control abstractions date to at least the 1980s [4].”
Reinforcement learning of motor skills with policy gradients (2008, methodology): “desired behavior ( Ijspeert et al., 2002, 2003 ).”
Imitation Learning (2017, methodology): “Alternatively, Ijspeert et al. [2002a] learn attractor landscapes from the demonstrations, creating a stationary policy that is at-tracted to the demonstrated trajectory.”
Incremental Semantically Grounded Learning from Demonstration (2013, background): “Dynamic Movement Primitives (DMPs) [9] are a formulation that can describe the evolution of dynamical systems over time using a system of nonlinear differential equations.”
STOMP: Stochastic trajectory optimization for motion planning (2011, background): “Goal-convergent exploration is highly desirable in trajectory-based reinforcement learning of point to point movements, where dynamical systems have been designed that satisfy this property [12].”
Towards Motor Skill Learning for Robotics (2007, methodology): “From this point of view, one can straightforwardly derive regression algorithms such as the ones in [4, 14] to imitate using both the standard formulation of motor primitives [4] as well as the perceptually coupled formulation [14].”
Neural learning of stable dynamical systems based on data-driven Lyapunov candidates (2013, background): “Another widely known approach is called dynamic movement primitives (DMP) [10] which is a very successful technique to generate motions with dynamical systems.”
GRAIL: A Goal-Discovering Robotic Architecture for Intrinsically-Motivated Learning (2016, background): “are dynamical models capable of producing whole discrete or rhythmic stable trajectories (see [33]), and efficient policy search algorithm to search the parameters of such policies”
Learning force control policies for compliant manipulation (2011, background): “(9) for both position and force trajectories.”