Control of Multi-Robot Systems in Intelligent Spaces
Recent advances in computing, networking, sensor technology and robotics
allows us to create a more convenient environment for humans. In that context
the Intelligent Space was proposed as a space that has ubiquitous distributed
sensory intelligence and actuators for manipulating the space. In that sense,
intelligent space is an environment with distributed sensors (e.g. cameras,
microphones, sonars) and actuators (e.g. mobile robots, manipulators), with
the purpose of providing various advanced services to the space users. The
sensors are used for the detection and tracking of objects and persons in
the space and for receiving orders from space users, and the actuators are
used for providing services to the users, both physical and informative (e.g.
carrying or delivering loads).
In context of using mobile robot as an actuator, in numerous applications
it could be more effective to use multiple simpler and cheaper mobile robots
instead of one complex and expensive robot. There are many concepts in control
of multi-robot systems, and we investigate the concept of multi-robot systems
in intelligent spaces. In terms of control and coordination of higher number
of mobile robots, the intelligent space offers several advantages compared
to the use of independent autonomous robots: a) it is not necessary to build
an environment map, b) in every time instance it is possible to uniquely determine
the position of the robot, c) even in dynamic environment, it is possible
to find globally optimal path, because the sensor system has always perception
of the whole environment, d) the whole intelligence is located in the environment,
instead in the robots, so that robots share common resources and this results
with lower system price.
On the other hand, sensors, both stationary and mobile, provide information
about the state of the space. For example, sensors are used to detect the
location and pose of humans and position of robots in the space. The availability
of this information enables the implementation of various personalized services.
The information from external sensors can also be utilized in mobile robot
control tasks, which gives an advantage in comparison with the standard approach
with only onboard sensors. In this topic, our research is mainly concerned
with global vision based robot position tracking algorithms, and also motion
planning and motion control algorithms, with the goal of cooperative performing
of more complex tasks.
Vision system for multiple mobile robots tracking
Our vision system consists of multiple distributed cameras fixed above the
robots, which track robots poses, and one or multiple distributed computers
connected via a high speed communication bus, which execute image processing
algorithms as well as decision making and robot control algorithms (see figure).
To achieve good precision of robot pose measuring, we designed a special
robot mark that consists of a color patch for detection, and a black-white
marker for measurement. This enables us to use subpixel edge detection for
measurement of robot pose. In this way we achieved outstanding measurement
precision with position standard deviation of about 0.007 pixels i.e. 0.03
mm, and angle standard deviation of about 0.08 degrees. Moreover, algorithm
is heavily optimized so that execution time to measure poses of 5 robots in
global search mode is about 3 ms. Where possible, algorithm uses advantage
of local search mode in which case the execution time to measure poses of
5 robots is only about 1.5 ms. Therefore, we are able to use high framerates
(currently we use 80 fps), which enables reliable and precise control of robots.
For detailed description of this vision system see our paper "Robust
and accurate global vision system for real time tracking of multiple mobile
Vision system for people tracking
We also investigate algorithms dedicated for tracking of humans inside intelligent
spaces, with special emphasis on reliable human detection and robustness to
illumination changes. Some initial results have been already accomplished
in this field, and now we aim to improve detection robustness.