Contest Event Chair: Lisa Meeden
Robots must enter a fallen structure, find and identify victims, possibly deliver a small package near the victim (representing a device to allow human rescuers to talk to the victim), possibly determine severity of injury, and help human rescuers determine the location of the victims. The robot must then exit the structure.
Robots will be judged in one of three categories based on basic ability to handle increasingly difficult scenarios. These are based on the part of the arena in which the robot competes, as described below.
The contest will be held in a structure (the arena) that allows simulation of various degrees of a collapsed building. On one extreme, a flat floor with minimal debris (large obstacles) will be available for those with indoor type robots who want to highlight other parts of the task, like planning and vision. On the other extreme, part of the structure will consist of a complete debris and rubble field that is most suitable for robots that can handle 3D maze structures. A middle section of the arena will give a transition from one extreme to the other, and will include at least two levels, with stairs and ramps of various slopes to get to the upper sections. In addition, robots will be presented with progressively smaller spaces in which to work. The sections also allow for heterogeneous teams of robots to transition from one section to another, so, for example, a larger robot might enter the easy section and then release a smaller robot that can then reach back into the smaller sections.
Although the sections represent a continuum from one extreme to the other, there will be clear transitions from the easy section to the middle section to the most difficult section. A team may chose to solve just the first section, or try to move to progressively harder sections.
The diagram below gives a general idea of the layout. The colors are just for referring to the drawing. The internal layout is only illustrative of what will be there; you can expect it to be different at the contest. The walls will be 4 feet tall, and 4 inches thick. They will be manufactured out of foamcore, but will be covered with normal interior finishes, such as wall paper and paint. Their may be doors. There will be windows in the walls. The yellow section is the easiest. The orange section is the medium difficulty area, which as described above will have multiple levels. The red area is the 3D maze and rubble field.
Click on the following links to find more detail on each section:
All sections of the arena, in addition to the victims to be saved, will have dangers to be avoided and to also be mapped to assist the human rescuers in avoiding them.
In all sections, the dangers include:
In the more difficult sections, the hazards will include:
The structure will not have any additional lighting, other than what is simulated to be coming from natural sources, or an occasional internal light source. The easy section can be assumed to have an adequate amount of light in the majority of its space. As the robots move into more difficult spaces, lighting may be nonexistent. Robots using the more difficult section of the structure should bring their own light sources, if needed.
Teams will be judged both for place awards, and for innovative technology. Place awards will be given for overall performance in the USAR task, with further place awards in each category of difficulty in the arena (assuming enough entrants to a particular category). The scoring is roughly equivalent to Olympic style scoring as used in recent AAAI robot contests. Each judge will have a certain number of points that can be awarded based on their measuring certain quantitative and qualitative metrics as follows.
Robots are judged in number of "human" targets found, and in time of finding the humans. Additional credit is given for accurately relaying information back to outside "rescuers" about the location of the victims. Additional disruption of the structure, for example, knocking over of large debris or causing additional collapses, will reduce a robot's score. Not being able to exit the structure before the time limit expires will also reduce a robot's score. (Note that ANY exit will do. The idea is that communications may not be reliable, and the robot might need to exit the structure in order to communicate the map to the human rescuers.) Credit will be given for identifying and mapping hazards. Credit will be given if a robot can deliver a small package to each victim. (The package represents water, food, and possibly a device to allow the rescuers to communicate with the victim.) Credit is also given for being able to communicate with a victim to determine the victim's degree of injury, and for quality of interaction with outside rescue teams (the researchers). Note that being able to work completely autonomously, locating victims and then returning from the structure to report victims locations and conditions is an ideal solution, but we are also encouraging mixed initiative systems. PLEASE NOTE: in the easy section of the arena, outside human controllers may not communicate with their robots. The mixed-initiative option is only available for the harder two sections of the arena!
FOR THE HARDER SECTIONS ONLY: If you will have any interactions with your robots during a run, you must be able to make it clear to the judges the level and quality of interaction. This might include textual output on the interaction (which then might be output to a speech synthesis system so that the judges and audience can hear the interaction), or visual output of interfaces that can be projected for the judges and audience. From a metrics point of view, the judges will be thinking like a "fireman under duress" when judging the interactions.
Another important capability when interacting with humans is what happens when communication is disrupted. If you can handle (and demonstrate handling) being tolerant to communications dropout, more credit will be given. (This will be considered additional difficulty in scoring.) If you want to be judged for handling dropouts, judges may simulate this by disconnecting your wireless ethernet, or by turning on a device to disrupt your communications.
Metrics in systems that use a human in the loop will include "commands" per minute and/or bandwidth used. Less commands per minute and less bandwidth per minute is more desirable.
Teams with more than one robot need to show the utility of cooperating or interacting robots, either in performing the task better, or performing the task more economically. That is, their team should have a better than linear time speedup, or be able to solve the task equally well in performance, but with better economic metrics, for example smaller, less expensive, lower power consumption, etc. As such, the scoring will factor in the number of robots, types of robots, types or mixture of sensors, etc., in determining a teams performance.
Each team will have a total time limit in which to find the victims, perform mapping, and exit the structure. The total time will depend on how far into the maze the team progresses. An initial time limit of 10 minutes will be given in the easy section. As the robot transitions into more difficult sections, additional time in 10 minutes increments will be added to the total time.
Only one team will compete at a time.
During competition rounds, the victims will be simulated. To simulate a victim, we will use a combination of clothing and a heating pad within the clothes (in addition to possible sounds and possible movement). For example, some victims might be loudly calling for help while others are softly moaning. The sounds will be produced by a tape recording. The heading pad is the "Dunlap Dry Heat Heating Pad model BHC112" available for about $9.00 at WalMart. The assorted clothes will most likely be brightly colored, but this is not guaranteed.
The victims will be designed to be inexpensive to assemble, and complete details on assembling a victim will be posted here in the near future to allow teams to develop sensors and practice.
The scavenger hunt contest, which was meant to be an abstraction of
problems like search and rescue, is still contained in a form somewhat
like the original, except that the contest is against time, and the
objects to be found will be in the easiest section of the USAR
structure. Also, the objects to be found will be those that represent
the human, and consist of color and shape, IR signature, and sound.
If you have a concern about the change in the scavenger hunt, please
let us know immediately!
We are very interested in showcasing and highlighting enabling
technologies in artificial intelligence. That is why part of the
arena is designed for larger, indoor robots. The simple part of the
structure has a flat floor that does not require an all terrain robot.
And the amount of debris is limited.
Sorry, but no. We DO allow systems with adjustable autonomy, or
mixed initiative, but this means that the human gives instructions
at a high level, and can work with the robot, but the robot must
have a degree of autonomy and able to perform some of the task
without a human in the loop. Please contact us if you are unsure if
your system would qualify.
We are under discussions with RoboCup-Rescue (including having a position on their steering committee) to coordinate these contests to allow better interaction of the researchers and to allow results from one to be easily applied to the other. We will continue to work closely to ensure the ability of teams to apply results from one contest to the other.