|
Abstract:
|
The concept of household robotic servants has been in our mind for ages, and
domestic appliances are far more robotised than they used to be. At present, manufacturers
are starting to introduce small, household human-interactive robots to
the market.
Any human-interactive device has safety, endurability and simplicity constraints,
which are especially strict when it comes to robots. Indeed, we are still far from
a multi-purpose intelligent household robot, but human-interactive robots and arti
cial intelligence research has evolved considerably, demonstration prototypes are
a proof of what can be done. This project contributes to the research in humaninteractive
robots, as the robotic arm and hand used are specially designed for
human-interactive applications.
The present study provides a distributed framework for an arm and a hand
devices based on the robotics YARP protocol using the WAMTM arm and the
BarrettHandTM as well as a basic modular client application complemented with
vision.
Firstly, two device drivers and a network interface are designed and implemented
to control the WAMTM arm and the BarrettHandTM from the network. The drivers
allow abstract access to each device, providing three ports: command requests port,
state requests port and asynchronous replies port.
Secondly, each driver is then encapsulated by YARP devices publishing realtime
monitoring feedback and motion control to the network through what is called
a Network wrapper. In particular, the network wrapper for the WAMTM arm
and BarrettHandTM provides a state port, command port, Remote Procedure Call
(RPC) port and an asynchronous noti cations port. The state port provides the
WAMTM position and orientation feedback at 50 Hz, which represents a maximum
blindness of one centimetre.
This rst part of the project sets the foundations of a distributed, complete
robot, whose design enables processing and power payload to be shared by di erent
workstations. Moreover, users are able to work with the robot remotely over
Ethernet and Wireless through a clear, understandable local interface within YARP.
In addition to the distributed robotic framework provided, a client software
framework with vision is also supplied. The client framework establishes a general
software shell for further development and is organized in the basic, separate robotic
branches: control, vision and plani cation.
The vision module supports distributed image grabbing on mobile robotics, and
shared-memory for xed, local vision. In order to incorporate environment interaction
and robot autonomy with the planner, hand-eye transformation matrices have
been obtained to perform object grasping and manipulation. The image processing
is based on OpenCV libraries and provides object recognition with Scale Invariant
Feature Transform (SIFT) features matching, Hough transform and polygon approximation
algorithms. Grasping and path planning use pre-de ned grasps which
take into account the size, shape and orientation of the target objects.
The proof-of-concept applications feature a household robotic arm with the ability
to tidy randomly distributed common kitchen objects to speci ed locations, with
robot real-time monitoring and basic control. The device modularity introduced in
this project philosophy of decoupling communication, device local access and the components, was successful. Thanks to the abstract access and decoupling, the
demonstration applications provided were easily deployed to test the arm's performance
and its remote control and monitorization. Moreover, both resultant frameworks
are arm-independent and the design is currently being adopted by other
projects' devices within the IRI. |
