Adaptive movement sequences and predictive decisions based on hierarchical dynamical systems

Abstract

Abstract-This paper addresses the question of how to create adaptive and smooth sequences of actions and how to decide among skill options in a continuous manner without the necessity of recurrent planning. Motion generation is based on serial and parallel blending of movement primitives (MP). MPs are modeled as dynamical systems on task coordinates with attractor behavior and augmented with additional signals to ease their coordination. Sequences and transitions between skills are realized in a unified way as bifurcating dynamical systems based on continuous-time recurrent neural networks. The neural output is used as activation signal for MPs. Besides continuous feedback from the controlled MPs, the neural dynamics is influenced by a cost term from a future prediction to allow the inhibition of an action flow that is expected to fail.First results are shown in a physical simulation environment on a high-DoF robotic hand-arm system. The system is capable of creating smooth transients of MPs. Robustness to disturbances can be observed as local adaptations of individual lowlevel MPs, flexible sequencing of MPs, and global error recovery by changing the whole strategy of how to perform a movement skill.

Learning control Lyapunov function to ensure stability of dynamical system-based robot reaching motions

“…Such a model represents a global map which specifies instantly the correct direction for reaching the target, considering the current state of the robot, the target, and all the other objects in the robot's working space. Such models are more similar to human movements in that they can effortlessly adapt its motion to changes in the environment rather than stubbornly following the previous path [6,7,8,9,10,11,12]. In other words, the main advantage of using DS-based formulation can be summarized as: "Modeling movements with DS allows having robotic systems that have inherent adaptivity to changes in a dynamic environment, and that can swiftly adopt a new path to reach the target".…”

Probabilistic progress prediction and sequencing of concurrent movement primitives

“…The drawback is, that the coordination is more difficult, as the system has more possibilities for activating the MPs. Luksch et al [6] introduced a framework where MPs can be activated concurrently without any restrictions. The MPs are coordinated using recurrent neural networks.…”

“…For the finger angles, the difference between state and goal is used, while for the orientations, the angle error is computed. The diagonal matrix Σ can be used to shape the feature around the attractor goal [6]. In general, our approach works with arbitrary features.…”

Learning to sequence movement primitives from demonstrations

“…For example, Luksch et al [12] model the system as a recurrent neural network (RNN) in which primitives can be concurrently activated and are able to inhibit each other. This RNN architecture leads to smooth movements of the robots.…”

“…The features are not global but assigned as output vectors to primitives, leading to one output vector x i per primitive p i . A more detailed view on our movement primitive framework can be found in [12]. Please note, however, that our methods are kept general and that they should be applicable to any movement primitive framework and feature set.…”