![]()
SECTIONâ1
PLENARY PRESENTATIONS![]()
INFRASTRUCTURE ROBOTICS: OPPORTUNITIES AND
CHALLENGES
GAMINI DISSANAYAKE
Mechanical and Mechatronic Engineering
University of Technology, Sydney, Australia
Current applications of robotics is distinguished from more traditional automation by the focus on machines that operate in relatively unstructured, difficult and often hazardous environments. The past decade has seen the deployment of a number of robotic systems in highly challenging application domains such as mining and agriculture. In particular, the potential safety, cost and health impacts from the use of robotics aids for periodic inspection and maintenance of civil infrastructure has led to a significant expansion of research activity in âinfrastructure roboticsâ. Two of the prerequisites for deploying a robot in the field are the ability to acquire and maintain a representation of the environment, and efficient motion planning. In scenarios where a machine and a human must collaborate to perform a task, an intuitive user interface, a joint understanding of abilities and joint management of task execution are also essential. This talk will summarise current research on these key competencies that underpin robot deployments in unstructured environments, with the focus on simultaneous localisation and mapping, and human robot collaboration. Future potential and key challenges in infrastructure robotics will also be presented together with a review of the research activities in this field at the Centre for Autonomous Systems, University of Technology Sydney.
![]()
COGNITION-INSPIRED ROBOT LEARNING AND CONTROL
JIANWEI ZHANG
Institute of Technical Aspects of Mulitimodal Systems
University of Hamburg, Germany
In a dynamic and changing world, a robust and effective robot system must have adaptive behaviours, incrementally learnable skills and a high-level conceptual understanding of the world it inhabits, as well as planning capabilities for autonomous operations. Future robot systems will benefit from the recent research on neurocognitive models in processing multisensory data, exploiting synergy, integrating high-level knowledge and learning, etc. I will first introduce multisensory integration methods for intelligent control of robots. Then I will present our investigation and experiments on synergy technique which uses fewer parameters to govern the high DOF of multifinger robot movement. The third part of my talk will demonstrate how an intelligent system like a robot can evolve its model as a result of learning from experiences; and how such a model allows a robot to better understand new situations by integration of knowledge, planning and learning. I will show some integrated results of operational mobile robot platforms with grasping facilities in a restaurant service scenario.
![]()
BIOLOGICALLY-INSPIRED MINIATURE JUMPING ROBOT:
FROM DESIGN TO CONTROL
NING XI
Department of Electrical and Computer Engineering
Michigan State University, East Lansing, Michigan, USA
Robots will transform our daily lives in the near future by moving from controlled industrial lines to unstructured environments such as home, offices, or outdoors with various applications from healthcare, service, to defence. Nevertheless, two fundamental problems remain unsolved for robots to work in such environments. On one hand, how to equip robots, especially meso-scale ones with sizes of a few centimetres, with multiple locomotion abilities to deal with the unstructured environment is still a daunting task. On the other hand, how to control such robots to dynamically interact with the uncertain environment for agile and robust locomotion also requires tremendous efforts. In this talk, I will present my research efforts to tackle these two problems in the framework of biologically inspired robotics. First, I will show how to use biologically principles found in nature to build efficient meso-scale robots with various locomotion abilities such as jumping, rolling, and aerial manoeuvring. Second, I will present a novel non-vector space control method for meso-scale robots which have limited computation power. The research in these two thrusts will pave the way for next generation bio-inspired, low cost, and agile robots.
![]()
UNDERSTANDING ANIMAL LOCOMOTION USING BIO-
INSPIRED ROBOTICS AND SOFT ROBOTICS
TIANMIAO WANG
School of Mechanical Engineering and Automation
Beijing University of Aeronautics and Astronautics, China
The current studies of fish-like aquatic propulsion are quite limited in their ability to control for physical determinates such as body shape, kinetic motions, flexural stiffness, skin surface scale and Mucus properties etc. Keeping one of these determinates constant while altering others in a controllable manner is impossible for the swimming live fishes. Bio-inspired robotic models offered the ability to manipulate and control individual physical determinate that affect the aquatic propulsive performance.
The research group at Beijing University of Aeronautics and Astronautics has developed bio-inspired robotics as well as multi-material prototypes to experimentally investigate several new topics of aquatic propulsion. A key aspect of the experimental setup is that the bio-inspired robotics and multi-material prototypes are self-propelled (thus to satisfy Newtonian equation of balance) and can swim against the flow in the lab water tank. While self-propelled swimming, measurement of speed, external forces and torques, internal power consumptions, cost-of-transport (COT) of the bio-inspired robot and the multi-material prototypes are synchronized with the motion program and high-speed video of the wake flow. Finally, pneumatically actuated soft robotics, that can efficiently grasp objects with different sizes and shapes, will be introduced in this presentation. This robotic device is completely soft and made of low-cost lightweight material. The grasping performance can be realized by very simple control without any feedback.
![]()
SECTIONâ2
ASSISTIVE ROBOTS![]()
A BEHAVIOR ADAPTATION METHODBASED ONHIERARCHICAL POMDPS*
YONG TAOâ
School of Mechanical Engineering and Automation, Beihang University, Beijing
Beijing, 100191, China
YOUDONG CHEN
School of Mechanical Engineering and Automation, Beihang University, Beijing
Beijing, 100191, China
DONG XU
School of Automation Science and Electrical Engineering, Beihang University, Beijing
Beijing, 100191, China
JIAQI ZHENG
School of Machinery and Automation, Wuhan University of Science and Technology, Hubei, Wuhan, 430081,China
This paper proposes a behavior adaptation method for companion robot based on the hierarchical partially observable Markov decision processes (H-POMDPs). The H-POMDPs have recently emerged as a powerful method for optimizing human-robot interaction (HRI) decision making in partially observable environments. The model is used for representing and learning HRI models for indoor elderly companion robot. This approach proposes an approximate solution by employing the H-POMDPs in the uncertain environments. The HRI states such as daily dialogues, news and whether broadcasting, motion speed and navigation-assist are used throgh the H-POMDP. The userâs speech command, touch-screen input, head position and body posture are detected as subconscious signals that indicate a userâs interaction preference. The experimental results show the effectiveness and feasibility of designing the H-POMDP behavior adaptation method. The H-POMDPs can train significantly faster than the original MDP. The H-POMDPs requires much less data, and can easily be extended to variables reducing the time and sample complexity.
1.Introduction
In the aging of the society, limited mobility and high dependence on others affects peopleâs self-confidence negatively. The current services for the elderly are insufficient. We need to find alternative ways of providing care to the elderly population, which will not only lower the costs, but also increase the comfort of living with the level of dignity that our elderly deserve [1].
For improving the living environment, life style and care model of the elderly through high tech equipment, many intelligent assistive devices such as intelligent wheelchair, smart nursing bed and companion robot, have been developed in the last decades. Companion robots are expected to communicate with non-experts in a natural and intuitive way. A companion robot will act in an unstructured environment, such as a private home or a rest home for old people, with people roaming around. Currently, there are no robots that are able to perform a combination of these tasks efficiently, accurately and robustly.
The interaction between service robot and human is a research focus and challenge. Especially, finding optimal strategies for human robot interaction operating in an uncertain environment is an active field of research in service robotics [2-5]. One of the major goals of HRI is designing as: robot systems that perceive their environment and execute actions. In particular, robot should be able to plan their interactive actions with human, so as to fulfill their task as accurate as possible.
The service robots are required to accomplish their interaction tasks based on perceptions about the environment and the userâs intention collected via the sensors. The decision making in the observable static environments has been solved effectively via the well-known Markov decision process (MDP) [6].The MDP solutions setup applies to a single adaptive interactive actions in a certain environment at discrete time steps. However, in most real-world domains, the robots are not able to discern the exact environment state and the userâs intention as the sensors are noisy and limited. The faulty and inaccurate observations are given with the sensor noise, and the robot is unable to observe the difference between states that cannot be detected by the sensor.
The companion robot sensor reading might require different action choices. In this paper, we consider the problem of how robot can cooperate to jointly accomplish a task, while facing the sensing limitations such as the capability to observe the state of the interaction. The case is studied wherein robots have to learn how to plan the actions, given uncertainty in the actions and sensor readings error. The need for HRI learning and planning is particularly pressing, given that...
