ADAPT: A Cognitive Architecture
for Robots
D. Paul Benjamin, Tom Achtemichuk, Oz Michaeli, Chonarop
Jamroendararasame, Siddtharth Patel
Pace University Robotics Laboratory
Damian Lyons, Mohamed Shaheen Ali
Fordham University Robotics Laboratory
Deryle Lonsdale, Rebecca Madsen
Brigham Young University Department of Linguistics
and English Language
ADAPT
Adaptive Dynamics and Active Perception for Thought
The goal of the ADAPT project is to create a robot that can
model its environment accurately in real time, and use that
model to perform tasks and interact with people using natural
language.
The emphasis in ADAPT is on solving problems by
reformulation (reperception).
The structure of the ADAPT architecture is based on linguistics.
We are building ADAPT by implementing the RS (Robot
Schemas) language in the Soar cognitive architecture.
ADAP T uses a sophisticated, multimedia internal world model.
Lessons Learned the Hard Way:
Everyt hing is sensory-motor.
Perception is an active process.
There is a high degree of true concurrency.
Analyzing in teractions is just too hard.
Active Perception
Active perception is top-down and goal-d irected, so
that perception becomes a problem solving proces s.
Perception is the process of constructing and
mod ifying intern al representations, so tha t
perception and repercep tion (re formulation)
becomes a central part of problem solv ing
(Newell, 1990) (Duncke r, 1947).
RS is implemented in Soar
RS
Soar
RS is a la nguage for real-time, distributed control of robots
with a formal semantics consisting of a hierarchy of port
automata. ADAPT is controlled by RS schemas.
Soar is a cognitive architecture that manipulates a hierarchy
of problem spaces. Soar provides the mechanisms for
synthesizing and learning RS schemas.
Predictive Vision
Tasks
“Shepherding” tasks: moving objects among enclosures
This invo lves ab stract task planning , motion planning,
navigation, grasping, perception
We can embed lot s of tasks commonly used in
cognitive science
These are inherently useful tasks.
1
2
Observer
3
Thanks!
Solving Problems by Decomposition
Reformulate the world into hierarchies composed of a small
number of recurring patterns.