News
September 2010: Shifted to lugano, Switzerland.
July 2010: Mobile Robotics Laboratory, IISc, is being renamed to Aerial Robotics Laboratory
:: Incremental Slow Feature Analysis ::
The Slow Feature Analysis (SFA) unsupervised learning framework extracts features representing the underlying causes of the changes within a temporally coherent high-dimensional raw sensory input signal. We develop the first online version of SFA, via a combination of incremental Principal Components Analysis and Minor Components Analysis. Unlike standard batch-based SFA, online SFA adapts along with non-stationary environments, which makes it a generally useful unsupervised preprocessor for autonomous learning agents. We compare online SFA to batch SFA in several experiments and show that it indeed learns without a teacher to encode the input stream by informative slow features representing meaningful abstract environmental properties. We extend online SFA to deep networks in hierarchical fashion, and use them to successfully extract abstract object position information from high-dimensional video.
(For more details and code please check out: IncSFA).
::
Collective Reward-Based Approach ::
An Image Based Detection of Semi-Transparent Objects
Figure 1: Sample input image and its corresponding final result.
Most computer and robot vision algorithms, be it for object detection,
recognition, or reconstruction, are designed for opaque objects. Non-opaque
objects have received less attention, although various special cases have been
the subject of research efforts, especially the case of specular objects. The
main objective of this work was to provide a research work in the case of semi-transparent
objects, i.e. objects that are transparent but also reflect light,
typically objects made of glass. They are rather omnipresent in man-made
environments (especially, windows and doors). Detection of these objects
provides important information that can be used in a robot's navigational strategies
such as obstacle avoidance, detection of oil/water spills on the floor, localization, etc.
In order to achieve the detection of semi-transparent objects
we developed a novel approach using a collective-reward based technique on
an image captured by an uncalibrated camera. Several experiments were
conducted over different scenarios to test the efficacy of the algorithm. Figure 1 : Images
Taken from the camera Left: Original
Image Left and right center: Painted images (Obstacle not detected)
Right: Painted Image (Obstacle detected)
Figure 1 :
VITAR-I For vision-based navigation
experiments, we have built a robotic testbed christened VITAR (Vision
based Tracked Autonomous Robot) that consists of a tracked mobile robot
equipped with a pan-tilt mounted vision sensor, an on board PC, driver
electronics, and a wireless link to a remote PC. A novel
appearance-based obstacle avoidance algorithm that uses histograms of
images obtained from a monocular camera is developed and tested on the
robot.
Figure 2 :
VITAR-II Vitar has evolved to its newer
version VITAR-II. A lot of mechanical modifications were made
to suit outdoor navigation. Vitar - II is more compact and light-weight
compared to its predecessor. Modifications to the mobile robot base and
design of more complex experiments are currently under progress. The
images of VITAR are available in the Gallery
section. The
glowworm swarm optimization (GSO) algorithm is an optimization
technique developed for simultaneous capture of multiple optimums of
multimodal functions and can be implemented in a collection of mobile
robots to carry out the task of multiple source localization. In this
work, we conduct embodied robot simulations by using a
multi-robot-simulator called Player/Stage that provides realistic
sensor and actuator models, in order to demonstrate the efficacy of the
GSO algorithm in simultaneously detecting multiple sources. The study,
based on embodied simulation experiments, also shows the robustness of
the algorithm to implementational constraints.
This
Robot performs a beacon-based docking operation. It is equipped with an
ultrasonic range sensor (SRF05) and a set of infrared receivers.
Docking system makes the robot to dock to a target at a specific
location and orientation. Two active IR beacons are placed in such a
way that the location of the docking target lies on the Voronoi
Partition of the beacons. These beacons transmit infrared signal in all
directions. The robot detects the beacons and successfully reaches the
docking-target with the required orientation. A detailed description is
available here. Videos
of the docking robot are available in
the Gallery
section.
Figure 3 : The Robot and the
Docking target
::
Histogramic Intensity
Switching::
with Dynamic Mask Allocation (HIS-DMA): Vision-based Obstacle Avoidance
Algorithm for Mobile Robots
We
introduce a new algorithm called the histogramic intensity switching
with dynamic mask allocation, a method
which helps a robot to avoid obstacles using vision as the only sensing
element. The
algorithm uses the histograms of images captured by a monochrome camera
to achieve
obstacle avoidance. Histograms with special masks on input images are
used to give rise to a switching phenomenon in intensities based on the
dominant regions of the masked image. The mask lengths are dynamically
determined by a method called the Dynamic Mask Allocation (DMA). The
method does not make use of any direct distance measurement of the
obstacles and indirectly captures the essence of the principle behind
time-to-collision
(TTC). The algorithm is tested in real time on
VITAR-II. Videos
of Vitar avoiding obstacles are available in the Gallery
section. 
::
VITAR
(VIsion based Tracked Autonomous Robot)
::
::
Robotic
Implementation of
GSO using
Player/Stage
::
::
Autonomous Docking System
for a
mobile robot
::
