Research Interests
Artificial Intelligence, Robotics, Intelligent Virtual Agents, Dynamical Intelligence Modeling, Hybrid Dynamical Systems, Automated Reasoning, Reasoning About Navigation, Cognitive Modeling, Applied Logic, Tumor Modeling, Eyetracking and Attention
CV/List of Selected Publications
My CV is available in PDF format.
Some selected publications, indicative of some primary research interests:
- Action selection and task sequence learning for hybrid dynamical cognitive agents
E. Aaron and H. Admoni.
Robotics and Autonomous Systems, 58(9), 1049-1056, 2010.
- Dynamic obstacle representations for robot and virtual agent navigation
E. Aaron and J. P. Mendoza.
Twenty-fourth Canadian Conference on Artificial Intelligence, LNAI 6657, 1-12, 2011.
- On the complexity of the multi-agent, multi-depot Map Visitation Problem
E. Aaron, E. Kranakis, and D. Krizanc.
Fourth International Workshop on Wireless Sensor, Actuator, and Robot Networks (Workshop, Eighth IEEE International Conference on Mobile Ad-Hoc and Sensor Systems), 795-800, 2011.
- Hybrid system reachability-based analysis of dynamical agents.
E. Aaron.
Innovative Concepts for Agent-Based Systems: Second
International Workshop on Radical Agent Concepts, LNAI 3825,
233-244, 2006.
Primary Research Focus: Adaptive Intelligence and Hybrid Dynamical Cognitive Agents
My research interests are centered on models and applications of adaptive, dynamically responsive intelligence in dynamic environments. My primary focus is designing and analyzing intelligence models for autonomous agents in complex environments, emphasizing hybrid dynamical system models--models that combine continuous and discrete system dynamics--and applications ranging from workplace courier robots to computer-animated guides through virtual worlds.
This research is motivated by goals for performance, design, and analysis.
- Agents' goals and behavior should adapt appropriately from experiences in their environments, such as a courier robot learning when two tasks it has been given are redundant.
- Agents should dynamically respond to immediate stimuli, such as a robot avoiding a collision with a moving cart or person in an office hallway.
- Agents' intelligence models should support clear design and rigorous analysis, for robustness and reliability in a variety of applications.
My hybrid dynamical cognitive agent research enables
dynamically responsive and adaptive intention-guided behavior for
goal-based systems, supports efficient collision-free navigation in
dynamic environments, and promotes the integration of levels of
intelligence that are often modeled separately--e.g., obstacle
avoidance and goal-directed action selection--in a formal,
unifying hybrid system framework for verifiable dynamic agents.
This framework can improve adaptation and overall performance, and it can
also support model checking-based system analysis of agent behavior.
Results from this research include several advances in adaptive intelligence and analysis:
- a new, dynamical system-based framework of dynamical intention for dynamically responsive, adaptive goal-directed behavior;
- a new reactive navigation method that succeeds at indoor navigation tasks that cause difficulties for other reactive methods;
- graph-theoretic analyses of robot networks and navigating agents;
- and model checking-based analysis of properties of navigating agents, such as a class of formalized, hybrid system-based metrics for the relative difficulty of dynamical navigation in various scenarios.
The Selected Publications noted above contain more information
about these research results and others.
Other Research Areas and Interests
My work also includes cross-disciplinary projects that connect to Cognitive Science or Computational Biology.
Cognitive Logical Inference, Student Modeling, and Eyetracking
Before turning my attention
to intelligent virtual agents and animation systems, I
developed formalized mathematical tactics for cognitive
inference modeling.
I concentrated in particular on developing a model of undergraduate students
carrying out logical proofs in a structured framework, but the ideas underlying
the work are not student-specific. The many facets of this work are described
in my dissertation, Tactic-Based Modeling of Cognitive Inference on Logically
Structured Notation. I have also written papers on its components, which
range from logic and formalized mathematics (Justifying calculational
logic by a conventional metalinguistic semantics) to cognitive science
and eyetracking research (Insight into theorem proving via eye movements).
I am currently further investigating the eyetracking component of that research. (See the Cornell University Computer Science 40 Years booklet, pg. 24, for a small sidebar column about my eyetracking research!)
There is more information in the data collected during my dissertation
research than has previously been analyzed, and my current research with
Barbara Juhasz (Wesleyan University Department of Psychology) is a
more thorough exploration of the ways in which eye movements can
inform our understanding of the cognition and modeling of logical
problem solving. More than 668,000 frames of data are encoded for our
ongoing data analysis.
Computational Biology and Tumor Modeling
Simulations of tumor development and treatment are often computationally
costly. I am working with Ami Radunskaya (Pomona College Department of Mathematics) and others to improve their
efficiency by developing modeling and simulation methods that incorporate dynamically sensitive variable scaling, enabling greater detail at critical areas without excessive detail at mundane tissue.
This application is strikingly similar to applications that arise in
both the system verification and intelligence modeling aspects of my
agent modeling work. In my verification research, efficient
yet accurate metrics of relative navigation difficulty may be
enabled by state space decompositions with a greater density of
states around critical locations (e.g., obstacles, targets) and a
sparser distribution in other areas. In my intelligence modeling research,
such variable scaling techniques can relate closely to models of
perception for animated characters: Because some element of the
simulation of an intelligent virtual agent possesses perfect world
knowledge, realistic characters must attend to only relevant stimuli
in the virtual world around them, requiring efficient focusing of
attention on specific points in space. This, in turn, involves
intelligently spending less processing time (perhaps none at all) on
irrelevant locations while performing detailed analyses on relevant
ones, which is the essence of my tumor modeling research.
My collaborators and I are presenting our method for tumor simulation and its behavior in various modeling contexts, and we then intend to extend this work with our variable scaling approach, demonstrating of how biologically inspired notions of practical simulation
equivalence can lead to new criteria for evaluating biological
simulations and, by extension, faster simulation methods.
Other Noteworthy Publications
- Modal Logic Semantics
One way to avoid difficulties with partial functions is by
employing the notion of underspecification. In modal
logic S5 and some semantically related logics,
underspecification preserves validity, so incorporating
underspecification into their semantics does not change the
classes of valid formulae. A formalization of
underspecification and results for these modal logics are
concisely presented in Formal justification of
underspecification for S5 (co-author: David Gries).
- The Monty Hall Problem
The notorious brainteaser "The Monty Hall Problem"
(also called "The Three Doors Problem") is the source of much debate.
Would people be more likely
to correctly perform the underlying mathematics if the puzzle were presented
in a frequency format rather than a probability format?
Would people be more likely to arrive at the correct answer?
Results are reported in Frequency vs. probability
formats: Framing the three doors problem
(PDF version) (co-author: Michael Spivey).
This paper is also extensively quoted on pages 149-151 of the book
The Monty Hall Problem (Oxford University Press, 2009).
- Nuprl
Most documentation about the Nuprl automated reasoning system
has been written from an "insider's"
perspective. I used the system but was not involved in its
development; I have a user's perspective. My A
User-Level Introduction to the Nuprl Proof Development System
(PS version) was solicited by the Nuprl
group at Cornell University.
Non-academic stuff
Music is one of my most important non-academic interests. I
have written about music (album reviews
and features), broadcast as a jazz DJ (a weekly
commercial radio show), and played in a few rock bands
that successfully made it out of the garage.
|
|
