Robotics

1
Robotics Biology Laboratory
Department of Computer Science University of
Massachusetts, Amherst http//robotics.cs.umass.ed
u Oliver Brock, TJ Brunette, Tom Buckley, Maxence
Crossley Filip Jagodzinski, Dubi Katz, Andrew
Roberts, Yuandong Yang
Lab Overview
Autonomous mobile manipulation with a degree of
autonomy and competency that allows operation in
unstructured and dynamic environments would
enable a variety of important applications such
as planetary exploration, elder care, and the
disposal of improvised explosive devices.
Computational rather than experimental techniques
are fast becoming the new microscope, allowing
us to understand biological processes in ways
that current biologists can only dream of. This
will lead to new to cures for many diseases.
Exploitation
Exploration
Many computational problems are difficult because
their solution requires searching a very large
space. We devise approaches that improve
performance by carefully trading off exploration
and exploitation. This allows us to focus on
those areas of search space that are most likely
to contain a solution. The techniques we are
developing permit us to discover efficient
solutions to increasingly difficult computational
problems in autonomous robot manipulation and
molecular biology.
Projects
Autonomous Mobile Manipulation
Molecular Biology
Motion Planning and Control
Protein Structure Prediction
Interactive Perception
Protein Docking
Protein structure prediction aims to determine
the 3-dimensional biologically occurring state of
a protein. Our technique, based on robotic
motion planning, exploits information acquired
during exploration to direct computational
resources toward the biologically relevant
regions. This dramatically improves the
efficiency of search.
The capability to interact with the environment
gives a robotic platform an important advantage
over conventional sensing devices. Instead of
being a passive observer, the robot assumes the
role of an active participant in its
environment.. Complex learning tasks such as
discovering the effects of gravity, friction, and
whether a doorknob is stuck can be simplified
when the agent can interact with its surroundings.
Protein docking aims to determine the binding
interaction between two molecules. Current
approaches assume the molecules are rigid.
Representing the protein as a kinematic robotic
arm allows us to leverage robotic algorithms that
reduce the conformation space explored during
docking, thereby allowing flexible docking.
Our motion planning methods do task-consistent
motion generation in unstructured dynamic
environments by exploiting workspace information.
To our knowledge, it is the first motion
generation method that meets the requirements of
autonomous manipulation.
Lab Resources
The Phantom desktop haptics device is used to
interact with proteins
URS is a comprehensive open source software
package for researchers in robotics, biology and
artificial intelligence.
UMan is a 10 DOF mobile manipulator with a
holonomic base and a dexterous hand. It contains
various sensors including lasers and cameras to
acquire data from the world and a strong
computational system.
A 176-core compute cluster with the majority of
the hosts being dual core 2.4 GHz Xeon processors
with 8GB of RAM.
Back Row TJ Brunette, Oliver Brock, Maxence
Crossley, Andrew Roberts Front Row Filip
Jagodzinski, Dubi Katz, Yuandong Yang, and Tom
Buckley
Funding