|
Towards an Omni-directional Orientation Manipulation of Spherical Piezoelectric Motor: A Torque Vector Composition Method |
|
Tai-Chi Hwang, Yu-Ting Sung(National Taiwan University of Science and Technology, Taiwan), Yu-Hsi Huang(National Taiwan University, Taiwan), Chi-Ying Lin(National Taiwan University of Science and Technology, Taiwan) |
| This paper presents the application of a torque vector composition method realizing omni-directional rotor orientations for a novel standing wave spherical piezoelectric motor. Composited vector analysis shows that the use of all stator pairs is able to offer omni-directional rotor orientations in the xy, yz, and xz planes. Experiments demonstrate the feasibility of the proposed torque vector composition method and the potentials of expanding omni-directional orientations to the whole 3D space. |
|
|
Real-Time Multi-Person Action Recognition with a Neural Compute Stick |
|
Young-Chul Yoon, Hyeonseok Jung(Hyundai Motor Company, Korea) |
| In this paper, comprehensive research process for practical multi-person action recognition is presented. We perform various experiments using 3D CNN considering both performance and time efficiency. Distinguished from previous studies, we consider a performance on an embedded platform which consists of an embedded computer, ZED2 camera and a neural compute stick. The neural compute stick has its own memory and can be utilized asynchronously. This is a pioneer work proposing a multi-person action recognition framework using a neural compute stick. |
|
|
Globally Optimal and Scalable Video Image Stitching for Robotic Visual Inspection of Electric Generators |
|
Leonid Kostrykin(Heidelberg University, Germany), Claus Rohr(Siemens Energy, Germany), Karl Rohr(Heidelberg University, Germany) |
| We introduce an image stitching method, which generates composite images of generator wedges from temporal videos using intensity-based registration and non-linear blending. In contrast to previous intensity-based registration approaches, our global method simultaneously exploits the information of all image frames of a video and directly determines the global image translations. We propose a suitable energy function and employ a graph-based method for globally optimal minimization in linear runtime. Regularization is used to exploit physical knowledge about the application domain which improves the robustness. We found that our method yields superior results compared with previous methods. |
|
|
Learning-Driven Exploration for Reinforcement Learning |
|
Muhammad Usama, Dong Eui Chang(KAIST, Korea) |
| Effective and intelligent exploration remains an unresolved problem for reinforcement learning. Most contemporary reinforcement learning relies on simple heuristic strategies which are unable to intelligently distinguish the well-explored and the unexplored regions of state space, which can lead to inefficient use of training time. We introduce entropy-based exploration (EBE) that enables an agent to explore efficiently the unexplored regions of state space. EBE quantifies the agent's learning in a state using state-dependent action values and adaptively explores the state space, i.e. more exploration for the unexplored region of the state space. |
|
|
Model-Based Reinforcement Learning with LSTM Networks for Non-Prehensile Manipulation Planning |
|
Jeffrey Fong(National University of Singapore, Singapore), Domenico Campolo(Nanyang Technological University, Singapore), Cihan Acar, Keng Peng Tee(Institute for Infocomm Research, Singapore) |
| In this paper, we propose an interactive learning framework that allows a robot to autonomously learn an unknown object's dynamics and utilise the learned model for efficient planning of non-prehensile manipulation. First, we model the overall object dynamics using a Long Short-Term Memory neural network. We then assimilate the learned model into the Monte Carlo Tree Search algorithm with a dense reward function to generate an optimal sequence of push actions for task completion. We demonstrate the framework in both simulated and real robot that pushes objects on a table. |
|
|
Partitioned Gaussian Process Regression for Online Trajectory Planning for Autonomous Vehicles |
|
Pavlo Vlastos(The University of California at Santa Cruz, United States) |
| Gaussian process regression and ordinary kriging are effective methods for spatial estimation, but are generally not used in online trajectory-planning applications for autonomous vehicles. A common use for kriging is spatial estimation for exploration. Kriging is limited by the necessary covariance matrix inversion and its computational complexity. Using the Sherman-Morison matrix inversion lemma, the complexity can be reduced. This work focuses on further improving the computational time required for spatial estimation with partitioned ordinary kriging (POK) for online trajectory-planning. |
|