RC@H 2011: 1st MICAI Workshop on
RoboCup@Home
MICAI 2011
November 29, Puebla, Mexico
The aim of this workshop is to share
experiences between the teams that have participated at robocop@home, and
promote collaborations among Mexican teams as well as with other teams in the
world. Topics of interest include but are not limited to:
- Robot platforms
- Map building, navigation and localization
- People detection and tracking
- Object detection and manipulation
- Human-robot interaction
- Robot software architectures
- Planning and coordination
Contact: Luis
Enrique Sucar, INAOE, Mexico
Participating Teams:
Golem (UNAM)
Markovito (INAOE)
Nanisha (UPAEP)
Pumas (UNAM)
Preliminary Schedule:
9:00 Opening
9:15 Teams - General Descriptions
Golem-II+, an implementation of the Interaction-Oriented Cognitive Architecture
Caleb A. Rascón Estebané (UNAM)
Markovito's Team Description RoboCup@Home 2011
L. Enrique Sucar, Eduardo F. Morales, Héctor H. Avilés, Patrick Heyer,
Pablo Oropeza, Manuel Oropeza, Hussein López, Irving Vásquez, David
Carrillo, and Roger Luis (INAOE)
Pumas 2011 Team Description Paper
Jesus Savage, Marco Negrete, Mauricio Matamoros, Israel Figueroa, Abel
Pacheco, Luis Contreras, Ismael Castillo, David Esparza, Francisco
Dorantes, Alejandra Sanchez (UNAM)
Donaxi: Service Robot
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella (UPAEP)
11:15 Break
11:30 Architectures and Competitions
Blackboard based Architectures for the Software of a Middle-end mobile robot
Mauricio Matamoros, Ismael Castillo, Israel Figueroa, Jesús Savage (UNAM)
Task Coordination for Service Robots based on Markov Decision Processes
L. Enrique Sucar, Eduardo F. Morales, Elva Corona, Héctor H. Avilés, Pablo Oropeza, Manuel Oropeza (INAOE)
Models and structure of the Robocup@home task using dialogue models
Lisset Salinas Pinacho (UNAM)
Robocup 2011 Tests: Follow Human and Demo Challenge
David Esparza, Marco Becerra, Marco Negrete, Israel Figueroa,Ismael Castillo, Luis Contreras (UNAM)
13:30 Lunch
15:30 Vision and Human-Robot Interaction
Hybrid Algorithm to Human-Face Detection and Tracking Unrestricted
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella (UPAEP
RGB-D Person Following for Human--Service Robot Interaction
H. Avilés, Patrick Heyer, L. Enrique Sucar, Eduardo Morales, David Carrillo (INAOE)
Vision algorithms in Golem-II+
Arturo Rodríguez García y Saúl Martínez Vidals (UNAM)
Identification of people on the floor due to a fall or other accident
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella (UPAEP)
17:30 Break
18:00 Panel
Collaborating and Competing Towards RoboCup 2012
20:00 Closing
Abstracts
Golem (UNAM):
Golem-II+, an implementation of the Interaction-Oriented Cognitive Architecture
Caleb A. Rascón Estebané
Abstract:
The Interaction-Oriented Cognitive Architecture (IOCA) focuses in a
human-computer interaction and human-robot model. This model is known
as a Dialogue Model. This represents the structure of the task that
both human and computer/robot will realise together. The Golem-II+
robot is based on IOCA, in this talk we will present this process. As a
case of study, we will present the Navigation Autonomous System of
Golem-II+ and general observations of its performance during the
Robocup@home 2011.
Models and structure of the Robocup@home task using dialogue models
Lisset Salinas Pinacho
Abstract:
In this talk we present a methodology to model the tasks of the
competition Robocup@home. The methodology is based on the concept of
the Dialogue Model. These are defined in terms of the expectations that
the system has for each of the potential situations that could
rise during the execution of the task. The dialogue model also defines
the actions to be execute once an expectation is reached. In the talk,
we will present a summary of the models for the competition
Robocup@home 2011, their success and failures.
Vision algorithms in Golem-II+
Arturo Rodríguez García y Saúl Martínez Vidals
Abstract.
The Golem-II+ robot has implemented a diverse set of computer vision
algorithms to perform the task on the Robocup@home competition. In this
talk, we will review the main algorithms and their integration to the
IOCA architecture using the concept of the Dialogue Model.
Additionally, we will present some results of the evaluations of such
algorithms in an independent and integrated fashion.
Markovito (INAOE):
Markovito's Team Description RoboCup@Home 2011
L. Enrique Sucar, Eduardo F. Morales, Héctor H. Avilés, Patrick Heyer, Pablo Oropeza, Manuel Oropeza, Hussein López, Irving Vásquez, David Carrillo, and Roger Luis
Abstract.
The idea of service robots to assist humans in every-day life has been
around for many years. Although there is much work in developing
different abilities for this kind of robots, little attention has been
paid for the integration of these behaviours into a complete functional
system. In this paper we present Markovito, a service robot that
considers a general and flexible framework to easily develop different
daily life tasks. We describe its hardware and software architectures,
its main modules and how these are integrated to perform the different
tasks in the robocup@home competition.
Task Coordination for Service Robots based on Markov Decision Processes
L. Enrique Sucar, Eduardo F. Morales, Elva Corona, Héctor H. Avilés, Pablo Oropeza, Manuel Oropeza
Abstract.
Markov Decision Process (MDPs) provide a principled framework for planing under uncertainty, so they are appropriate for task
coordination in service robots. We have developed a methodology to
coordinate the different modules of our robot, Markovito, based on
MDPs. An MDP is defined for each task, and solved to obtain an optimal
policy. This policy is used to command the different modules through
high level actions, which in this way are coordinated to complete the
task. In this way we can easily re-use the same basic capabilities to
perform different tasks according to their goal, and at the same time
consider the inherent uncertainty present in real environments. We have
used this framework for task coordination for Markovito, we present the
MDP models developed for different task in robocup@home. We will also
discuss our recent work on using DEC-MDPs so a complex
task can be decomposed into several MDPs which are solved independently
and their solutions are combined. This facilitates the construction and
solution of the MDPs, and allows concurrent actions.
RGB-D Person Following for Human--Service Robot Interaction
H. Avilés, Patrick Heyer, L. Enrique Sucar, Eduardo Morales, David Carrillo
Abstract.
We present a novel visual system based on RGB-D camera mounted on a
service robot to effectively track and follow a user in indoor
environments. Our system discretized coarse depth data into five ranges
of a fixed length, and segments homogeneous regions of interest using a
simple and fast scan line algorithm. Motion commands of the robot are
constantly decided accordingly to its distance and position with
respect to the user. This approach was tested in the 2011 Robocup@Home
competition with competitive results, that show the suitability of this
proposal and the ability of our robot to follow users in real world
scenarios.
Pumas (UNAM):
Pumas 2011 Team Description Paper
Jesus Savage, Marco Negrete, Mauricio Matamoros, Israel Figueroa, Abel
Pacheco, Luis Contreras, Ismael Castillo, David Esparza, Francisco
Dorantes, Alejandra Sanchez
Abstract
This paper describes the team PUMAS from the University of Mexico,
UNAM, this team had participated in the RoboCup@Home since the first
competition in Bremen in 2006, then in Atlanta in 2007, where our team
obtained the third place, we also participated in China in 2008 and
also in Graz, Austria in 2009, where our team qualified for the second
stage of the competition, and in Singapore in 2010. Our robot uses a
robotics architecture, the ViRbot, used to control the operation of
service mobile robots. It accomplish the required commands using AI
actions planning and reactive behaviours with a description of the
working environment. In the ViRbot architecture the actions planner
module uses Conceptual Dependency (CD) primitives as the base for
representing the problem domain. After a command is spoken to the
mobile robot a CD representation of it is generated, a rule based
system takes this CD representation, and using the state of the
environment generates other subtasks represented by CDs to accomplish
the command. By using a good representation of the problem domain
through CDs and a rule based system as an inference engine, the
operation of the robot becomes a more tractable problem and easier to
implement.
BLACKBOARD BASED ARCHITECTURE FOR THE SOFTWARE OF A MIDDLE-END MOBILE ROBOT
Mauricio Matamoros, Ismael Castillo, Israel Figueroa, Jesús Savage
Abstract.
As hardware evolves, new and more powerful devices and processors
become available for robotics, so the software which controls those
robots must evolve too. Now days, most middle-end mobile robots (those
which human-like capabilities) are controlled by a software based on a
Peer-to-Peer architecture which lacks on changeability and reliability
in order to achieve higher performance. Our proposal opposed to this
distributed tendency is the use of centralised Blackboard-based
software architecture to get a more changeable and reliable software
without a negative impact on performance when the system’s granularity
is middle or coarse.
Robocup 2011 Tests: Follow Human and Demo Challenge
David Esparza, Marco Becerra, Marco Negrete, Israel Figueroa,Ismael Castillo, Luis Contreras
Abstract.
This work is a case of study of how several separate systems of
software are coordinated in order to accomplish two separate tasks:
following a human and cleaning a table, as happened in Robocup 2011.
When the robot gets an order to perform that task, the Action Planner
(an expert system programmed in CLIPS) calls the appropriate
instructions to carry out the task; later, the Simple Task Planner
module takes control of low level behaviour modules and coordinates
them, besides of generating information relevant to the robot when
necessary. Each low level behaviour module executes instructions
related to a piece of hardware, such as the
mobile base, the Arm or the Head, the laser sensor or the RGB-D camera.
The person tracking system makes use of a laser sensor to detect and
track the legs of a person, so the robot can follow him or her when the
person is walking. The vision system makes use of the capabilities of
RGB-D cameras, which both provide visual information (images) as a
conventional camera and the depth of every element (pixel) in the
image. Pumas@Home team makes use of the capabilities of these cameras
with novel algorithms for object detection, recognition and
localisation.
Nanisha (UPAEP):
Donaxi: Service Robot
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella,
Abstract.
Donaxi is a service robotics in domestic activities. The robot Donaxi is used
as a vehicle for research in control of movement and human-robot interaction. The at
Home League of RoboCup provides an ideal tested for such aspects of dynamic in
motion indoor, skills to manipulate objects, aptitudes to intergesticulate with the persons
in natural language and more. Donaxi has participated in RoboCup 2009 and RoboCup
2010, has gradually improved some skills and has changed its structure to better interact
with their environment in the RoboCup tournament. A modular software architecture as
well as further technologies have been developed for efficient and effective
implementation and test of modules for sensing, planning, behaviour, and actions of
service robots. To face and object recognition we use the OpenCV, and others
approaches and resources.
Hybrid Algorithm to Human-Face Detection and Tracking Unrestricted
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella,
Abstract.
The tools used for face detection and tracking are: face detection, face
recognition, contour of face detection, face detection position, a priori knowledge of
the next transition state for tracking, a priori knowledge of the contour of the face
without detection, processing time interval and additional features of the dress of the
person. The idea of the algorithm is as follows, the situation is normal when there is a
face in the image and corresponds to someone you know, so we can label and follow
the sequence of images with a priori knowledge that the face does not disappear or
change position drastically, that is, remains in the vicinity and if it disappears due to
small perturbations in that vicinity. When the person turns around, we know that the
boundary in that vicinity belongs to the person recognised and so we can follow the
contour, ensuring with some other features of the dress of the person. Clearly, there
are many processes running concurrently so that execution times can be altered,
however we have an estimated processing time for the proposed algorithm, for which
the parameters can be altered if the processing times change, it is where we must
change the behaviour of face tracking.
Identification of people on the floor due to a fall or other accident
Hector Vargas, Monica Lopez, Edson Olmedo, Gerson Beristain, Enrique Sanchez, Vittorio Zanella
Abstract.
The elderly people are routinely fall for normal atrophy of muscles and
bones and if not assisted promptly may have serious complications. Unfortunately, a
considerable population stands alone at home, this leads to non-assisted promptly and
thus suffer serious injuries. One solution is that the person has a service robot to assist
him in any situation, particularly when suffering from a fall. The robot must know
that there a person on the floor and background knowledge the robot determines that
it was due to a fall, and assisting the person with two activities in parallel, that is, the
robot tries to talk to the person to determine the health status and to send a message
with a short report to the doctor and when the robot has more information sent. For
the service robot can determine that it is a person on the floor and was due to a fall or
accident, you need to consider the following: associating data to the virtual map (a
priori knowledge of what was in that place ), seeking a face, recognise faces, to
recognize features of the dress of the person, locate a point of care, talk to the person
and send the person's condition.