Continuaremos os seminários não realizados da semana passada sobre planejamento simbólico/
geométrico em caso de desastres e gerenciameto de recursos para robótica móvel embarcada.
Após cada apresentação haverá discussões sobre os temas abordados.
Data: 15/05/2018
Horário: 14h00
Local: P32, sala 516
Agenda:
=======
Apresentador: Renan Guedes Maidana
Título: Autonomic Computing Towards Resource Management in Embedded Mobile Robots
Abstract:
Small and lightweight robots are becoming a reality due to the increasing pervasiveness
of embedded computers. However, much of the research in mobile robotics is focused on
increasing the robots' autonomy, disregarding the computational limitations of embedded
computers. The area of autonomic computing defines self-managing systems, also known as
autonomic systems, which manage computational resources (e.g. CPU load) and reduce the
effect of processing bottlenecks, as well as alleviating energy consumption in embedded
mobile robots. In this paper, we propose the use of autonomic systems with control theory
for resource management in embedded mobile robots. Our expected contribution is to use
autonomic computing to develop self-contained embedded robots, capable of operating with
few embedded computers and without the aid of external servers. We implement an autonomic
system which successfully controls the latency and CPU load of a dense optical flow appli-
cation, running on a Raspberry Pi 2 embedded computer, which validates the proposed idea.
Duração (estimativa): 45min
Apresentador: Mauricio Cecilio Magnaguagno
Título: Symbolic-Geometric planning for disaster/rescue situations
Abstract:
Classical planners, with actions described with preconditions and effects, create a way to
operate on (mostly) purely symbolic models in order to find plans to reach an agent's goals.
Plans found by classical planners often lack the geometric and temporal details required to
solve motion problems, such as grasping an object or avoiding collisions with moving objects.
Such details include robot dimensions and object properties.
A possible approach is to mix off-the-shelf classical and motion planners and share data
between the parts, which generates a large search-space without significant relations between
the objects. In order to tackle this problem, we need a custom symbolic-geometric planner to
share and limit possible values as planning progresses, to minimize memory usage and planning
time.
To better limit which data must be considered we can describe search paths as strategies to
force the planner to consider certain user preferences to obtain a faster solution with a
better plan quality. This is specially important in emergency response planning as rescue
teams may include people that already have certain methods to operate, and robotic agents
(new team members) must match an expected behavior to increase the efficiency of the team.
In this seminar we explain the advantages of HTN planning over classical approaches, how data
can be shared between the symbolic and geometric parts while considering temporal constraints
described to the system, and some applications in disaster/rescue situations.
Duração (estimativa): 45min
