AAAI 2006 Scavenger Hunt Event
Paul Oh
Drexel University
paul@coe.drexel.edu
Zachary Dodds
Harvey Mudd College
dodds@cs.hmc.edu
What's New
6/15/06 The Scavenger Hunt Schedule
has been posted. So have a draft of the Judging Guidelines.
6/1/2006
We thank K-Team and
Road Narrows Robotics
for donating a Korebot Light Linux SBC robotics board to the
scavenger hunt event, to be awarded to the overall first-pace team.
Overview
In this year's scavenger hunt robot competition, entries will
reason about and
navigate the conference's foyer, hallways, and rooms. We welcome
entries from any subfield of AI with a spatial reasoning component.
Systems incorporating spatial-reasoning techniques from disparate
subfields, e.g.,
natural language processing, human-robot interaction, multiagent
cooperation, and/or sliding-scale autonomy are encouraged, as are
more traditional entries focusing on navigation and mapping.
The competition will consist of two phases: a demonstration created
primarily by each participating team and a challenge directed
by the contest judges.
The demonstration phase
In the demonstration,
participants will show off their system's
abilities within the conference environment. For example, a particpant
might wish to demonstrate the ability of a robot to follow a trail of
colored paper in the conference hall, receive a designated visual cue,
and then head to an "X" marking the spot of their treasure. Another entry might
create a spatiotemporal map of the people around it and their
interactions. This might be accomplished through sensor observations or
by interacting with passersby. During this demonstration phase,
particpants largely set their own goals and exhibit what their system
is capable of. This phase of the competition is especially friendly
to other robotic contest entries, educational projects, and
systems that take novel AI approaches to environmental reasoning.
The challenge phase
In the predetermined challenge, robots will have to identify
the location of a number of objects chosen by the judges ("the hunt").
These objects will be selected to respect the primary
sensing modalities of each entrant: generality is an objective,
but not the only one. Examples of some of the possible objects appear
below. This task requires robots to explore a dynamic area, including
moving objects/people, in order to acquire objects to satisfy the
checklist.
In addition to this scavenger hunt challenge,
judges may ask participants to vary the operating conditions
from their demonstration phase in order to explore the limits of
their system's capabilities.
Each contestant will be given a list of items to find. In order
to be as concrete as possible, these items will be chosen
from the page linked FROM THIS SCAVENGER
HUNT ITEM PAGE
- all are available at very low
cost from Wal-Mart or other similar places.
System-specific Accommodations Welcome
To encourage all kinds of entries,
participants may substitute the above items with others,
if the sensor suite of their robotic system requires it.
Accommodations will be worked out on an individual
basis - teams are welcome to bring their own "objects"
to find during the challenge phase in any case.
In general, system-specific alterations
to accentuate the capabilities (and downplay limitations)
of each entry are welcome.
For example, an entry with range sensors but without a camera
might substitute distinguishable architectural features,
such as a particular corridor or niche, for the above items.
A system using audio input might seek out the source of
a particular voice or recording being played. Systems
whose focus is human interaction might seek out a particular
individual or, more generally, a person exhibiting a specified
behavior.
Regardless of approach, robots should report the location of
the scavenger hunt items they
find. This report may be in the form of a natural language utterance,
a map of the environment showing the location of items, or if the item
can be manipulated, by picking up the object and returning it to the
starting point.
Contestants may enter a team of robots and will be more favorably
judged if they demonstrate some form of cooperation.
Scoring
We recognize that direct comparison of
potentially very different entries is not easy.
However, the judges will base their fundamental assessments on
the extent and success of the spatial reasoning
that each system demonstrates, given the naturalness and
perceived difficulties of its operating conditions.
Minimum requirements are a mobile robot.
Each judge on the panel will give a subjective score
between 1 and 10 for each entry.
Their scores will be averaged to produce a final score.
The following criteria
will be used as a guide for the judges' considerations.
- Autonomy and shared autonomy
We welcome a variety of teams to enter with one or more robots and/or
human operators, though every entrant must demonstrate AI techniques
during the competition.
In particular, we encourage urban search and
rescue teams with AI components to consider joining this event.
Approaches resulting in systems with shared autonomy or full autonomy
will be considered on equal footing. In shared-autonomy
systems, judges will consider the naturalness of both the interface and
the delegation of tasks to the robotic system and its human
assistants. In fully autonomous systems, the extent of that autonomy
will be evaluated.
- Environmental modification
Ideally, an entry would interact with the conference
environment without modification, or by modifying the
environment itself. By default, the staging area for the
scavenger hunt will be a foyer of the conference venue, along
with its accompanying hallways and rooms.
The environment will not be engineered for the event, except that the
density of people will be relatively low, so that crowding
around a robot will not be allowed.
Participants are also welcome to
demonstrate capabilities under restricted conditions. In
such cases the nature and extent of the restrictions
should be well understood and conveyed to the judges.
- Unexpected, dynamic, and/or human interactions
A key aspect of the scavenger hunt competition is having
robots interact with people present in the environment.
This category will assess systems' ability to handle unmodeled
activity or changes in the environment. Robustness to such
phonomena is a hallmark of intelligent spatial reasoning.
As with the other judging criteria, participants may request
onlookers and judges to keep to specific types of interactions.
Robotic systems which make such requests for themselves will
be judged even more favorably.
- Accuracy
In order to convey its reasoning about the environment, each
scavenger hunt entry should create and convey one or more representations
of its surroundings.
Many such "maps" are possible, e.g., traditional dense maps,
sparse, loosely-connected collections of landmark locations,
networks of learned parameters,
or other summaries of the systems' spatial input data. Novel
representations or approaches integrating diverse facets of AI
are welcome. Judges will consider
both the accuracy and utility of these representations
in the demonstration and challenge phases of the competition.
- Range and completeness
Judges will assess the subset of the conference environment
which each system can cope with, especially in light of the
particular sensors available to each entry. For example,
a system equipped with a laser range finder would be expected
to reason about a larger swath of area than one with only a
set of IR sensors. "Completeness" considerations include
the variety of sensory modalities supported and their extent.
For example, can a system locate objects not on the floor?
Can a system distinguish objects using visual, auditory,
direct range-sensing, or other means?
- Speed
is desirable, but it is not as important as a system's
ability to interact with and reason about the (relatively)
unmodified conference environment.
The contestants will be
evaluated on overall success as well as on any particular abilities
they incorporate into their solutions, such as:
- Technical Innovation
- Novel spatial-reasoning approaches
- Mapping and navigation strategy
- Object Recognition
- Object Manipulation
- Multi-Agent Cooperation
Environment Details
Linked here is a pdf showing the atrium area of the
conference venue - this is where the robot
competition and exhibition will be held,
including the scavenger hunt.
atriumLayout.pdf
Prizes
A overall first place will be determined. Additional places and
prizes for innovative aspects of specific solutions may also be
awarded.
AAAI Robot Competition Home Page