BENG, M William (2018) Black Box Optimization of Dynamic Movement Primitives PFE - Project Graduation, ENSTA.

[img]PDF
56Mb

Abstract

The German Aerospace Center (DLR) developed a disruptive robotic system that aims to get closer to human capabilities in terms of dynamics, dexterity and robustness called David (formerly the DLR Hand Arm System). David has joints with Variable Stiffness Actuators (VSA) that have mechanically adjustable flexibility in the drive train. Because of its complexity, to make full use of this system new control schemes have to be found. Moreover, the typical approach to robotic grasping consists of the following steps: estimate the pose of the object, compute the goal configuration, use a motion planning algorithm to generate a collision free trajectory from start to goal configuration and finally track this trajectory with a controller. The motion planning usually has to be repeated at every grasp attempt and can be computationally expensive, especially for complex robot models, and generate unnatural movements. One alternative is to include learning and make use of demonstrated movements with Dynamic Movement Primitives (DMPs). These learned movements can later on be optimized with respect to a goal task. In this project, DMPs were used to generate movement primitives on particularly complex robotic systems. This framework was first implemented in simulation using a model of a humanoid robot to make a sit-to-stand motion much faster, then on a model of a robotic arm to make a arm-lifting movement as energy efficient as possible. DMPs were then implemented on the real David system. Energy optimization of an arm-lifting movement was tried out using the actual electrical power consumption measurement of the motors, but we realized that these measurements were not precise enough for the optimization to be carried out. Henceforth, in order to highlight the advantages of David's VSAs in terms of dynamics, we applied this framework to optimize a grasping task.

Item Type:Thesis (PFE - Project Graduation)
Subjects:Information and Communication Sciences and Technologies
Mathematics and Applications
ID Code:7196
Deposited By:William Beng
Deposited On:27 nov. 2018 10:56
Dernière modification:27 nov. 2018 10:56

Repository Staff Only: item control page