In order for robotic devices to perform useful tasks they
usually need to execute sequences of motions. I am interested in
the simultaneous computation of task plans and motion plans: a
discrete sequence of actions and its corresponding continuous
motions.
In practical applications of motion planning only noisy sensor
data is available. I am interested in reducing the gap between
perception and planning so that sensed data is more quickly
considered in the motion planning process and planning data is used
in the perception process.
I am interested in the development and application of motion
planners that consider the physical properties of robotic devices,
and allow fast computation times at the same time.
Estimating Coverage in High-Dimensional State Spaces
Estimation of coverage while searching high dimensional spaces
is a common problem for many sampling-based motion planning
algorithms. I am interested in the automatic computation of lower
dimensional Euclidean projections that allow estimation of
coverage.
Motion Planning with Real Perception on a PR2 from Willow
Garage. The motion planning code used here is part of the OMPL library, unless otherwise
specified.
A demo of sampling-based motion planning combined with a
trajectory optimization technique (CHOMP) to safely approach and
manipulate objects in a replanning context with real-time
perception.
Manipulation of known grasped objects in a complex environment.
Manipulation of known grasped objects in a complex environment.
Moving a glass. The task is to move the glass to the right for
30cm without spilling its content.
Moving a glass. The task is to move the left-most glass 50cm to
the right, without spilling its content.