Rules for the NCAI ROBOTICS COMPETITION July 11--16, 1993 Washington, D.C. Draft of March 17, 1993. Mobile robotics is an area where much of the research in diverse AI areas can be effectively and creatively combined to give interesting results. Last year's competition was a three-stage event in which mobile robots demonstrated skills of reactivity, exploration, and directed search (a detailed description is in the Summer 1992 issue of AAAI Magazine). At MCAI93, we would like to extend the competition to cover as wide a variety of mobile robots as possible, and to stress the ``intelligent'' aspects of their behavior. There will be three events. Each team can enter as many events as they choose, and may use different hardware and software platforms in the different events. There will be prizes for each event, as well as an overall winning team. We also encourage participation by those wishing to display innovative robotics designs without entering the competition. The focus of the events is mobile office robotics. This is an important application area for mobile robots, as well as a familiar one for the majority of mobile robotics researchers in universities and private laboratories. Given current hardware and software limitations, the tasks and environment will not be totally realistic, but we will try to make the events push the current state-of-the art. To encourage innovation, events are specified as tasks to be done, rather than giving rules on how it is to be done. GENERAL Robots. Last year there were separate arenas for large and small robots. The size and speed differences between these robots were not great; the largest robot was the Cybermotion K2A platform with a one-meter diameter, and the smallest were Nomad-class platforms at about 1/2 meter. All robots will compete using the same arena. There are no small-robot (< 20cm diameter) entries; there have been a few comments that the current version of the rules discourages them, and perhaps next year we should make special arrangements for them. The largest robot that can be entered must have a diameter less than 1.2 meters and height less than 2 meters. Thinner robots may have some advantage in negotiating obstacles, as some paths may not accommodate larger robots. However, we will make sure that all critical entries (doorways, for example) are large enough. The arena area will vary for the events, from approximately 10 by 15 meters to 15 by 23 meters. Arena outlines for the events, as well as typical setups, can be found in the three Postscript files: eventN.ps. Multi-bots, or groups of cooperating robots, can compete as a single entry. They will compete in the same events as single robots. See the writeup of individual events for special starting and finishing conditions. Robots must be powered by an on-board electrical source. They must have a visible, easily-operated "kill" switch that quickly stops any motion. Unlike last year's competition, there will be no separate event for testing whether a robot is "housebroken" or not. We expect robots to exhibit reasonable obstacle avoidance behavior, and not to damage the arena or create a danger to the audience. Walls, floors and objects. Like last year, the walls will be panels of a plastic material approximately 1 meter high. The sonar and infrared reflection characteristics of these panels are surprisingly good. We do not yet know what the floor will be like, but it will probably be wood or concrete rather than a soft rug; at any rate we will make sure that it is appropriate for wheeled robots. The color and composition of the floor will be sent out when it is known. Lighting conditions of the arena are also unknown at this time. Objects for the different events will be ordinary office furniture and cardboard boxes of various sizes. The boxes are the ones that were used at last year's competition. They are of standard cardboard construction; closed on all sides, top, and bottom; approximately 1 meter high, and of various sizes depending on the event; typical size is 1 meter square. No dimension of the boxes will be shorter than 50cm or longer than 1.5 meters. The color of the cardboard is off-white, and there will be contrasting dark tape on their edges, about an inch on a side. All boxes will be empty, and should be light enough for Nomad-class robots to push around, although this will depend on the composition of the floor. There will be more information on the weight of the boxes when it is known. Communication. No tethered robots will be allowed. In last year's competition, there were many problems using radio links for digital and video communication with the robot. This will probably be true this year also. There will be no special allowances for communication problems; teams are encouraged to use on-board computation for all critical functions. There will be side-by-side events, in which two robots will be competing simultaneously (although they will always be physically separate); this may also cause problems for radio communication. Although last year's rules allowed for communication from team to robot, almost noone took advantage of this. The result was that all robots functioned completely autonomously. This was a Good Thing, and one of the main points of the competition. This year, we will still allow limited forms of advice to be given to the robot, for example, if it gets hopelessly lost. There will be some penalty for doing this, depending on the amount of information conveyed. Unfortunately this is a hard point to judge objectively. As teams tell us the type of advice they expect to use, we will negotiate the penalty. Another aspect of communication is educating the audience about the technology of the robot systems. We strongly encourage this, to make the competition and exhibition a learning experience. There will be monitors in the competition area to display computer graphics or videos. Teams may submit videos describing their projects, and also may display the internal workings of the robot during its execution. Voice communication by the robot during its activities, giving information as to what it is doing and why, is especially effective. We will try to have a wireless mike available for broadcasting from the robot. We will also try to have some kind of overhead projection equipment that can be run directly from a high-resolution computer screen. Extra points will be awarded for effective communication. Perception. Sensor hardware and processing is one of the most difficult areas for most teams. Last year, teams were allowed unrestricted use of perceptual markers in the environment, and could design high-performance, special-purpose sensor systems using these markers. This year we would like to transition to a more "natural" environment, while still recognizing the difficulty of the perception problem. Given the ubiquity of visual cues in human environments, we will include such cues in each of the events. The type of a box, when appropriate for an event, will be indicated both by a visual marking on at least two sides of the box, as well as the shape of the box. Examples of all types of perceptual markings can be found in the Postscript file marks.ps. Details on placement of the markings can be found in descriptions of the individual events. Visual cues place a premium on vision sensors, although laser-light ranging systems and, to some extent, sonars can take advantage of shape cues. To make perception easier for teams without sophisticated sensor systems, we allow team-supplied perceptual markers to be used. The number and placement of these markers is limited according to the event, and there is a penalty for their use. Although noone did this last year, robots may place perceptual markers themselves at no penalty. Examples include dropping objects, marking walls, etc. Placed perceptual markers must be passive, that is, must not radiate energy on their own. Please try not to be messy: we will ask teams to clean up after themselves. A Priori Information. This includes general information about the environment, as well as particular configurations of walls and objects for each of the events. Teams can take advantage of whatever information is available from the competition description. The amount of information supplied varies with the event; see the descriptions below. We recognize that the performance of robots is heavily dependent on advance knowledge of many aspects of the environment, and if some points are unclear, teams are encouraged to ask us to clarify them. However, teams should also be prepared for variability in the environment. The diagrams furnished as typical arenas will be subject to some minor changes in sizes, shapes, and orientations. So will the perceptual markers and their placements. We want to encourage teams not to rely too strictly on absolutes about the environment. EVENTS Event I: Escape from the office! A typical configuration is in the Postscript file event1.ps. The arena area will be 10 meters by 15 meters. The office is approximately 4 by 5 meters; this size may vary somewhat. It will be against the wall farthest from the finish line, approximately centered along the wall. There are three "doors", described below, in the office, one on each of the walls that are not part of the arena. The exact location of the doors may vary. Robots will be started in the enclosed office area, in an arbitrary location but facing the finish line. Sometime between 1 minute and 3 minutes from the start, one of the doors will be opened, and the robot must find its way out into an open area and across a finish line. There will be obstacles (boxes) scattered throughout the rest of the arena. No prior knowledge of obstacles is given; they may be in various configurations, including ones that form concave shapes. No object should be moved or bumped; touch sensors are allowed as a means of sensing boxes, but should not move the boxes themselves. Points are awarded for speed and safety (not damaging objects or the robot). Maximum time limit is 15 minutes from when the door is opened. The door will be a panel in the office wall, 1.5 meters wide. There will be the same visual and shape cues on each of the doors. The visual cues are crosses ("+" signs) of contrasting color. Two of these will be placed at the same arbitrary horizontal position on the door, one at the top and one about 6 inches off the floor. The shape cue will be a raised strip 2 inches wide and 2 inches high running vertically from top to bottom at an arbitrary location along the door. Teams may add their own perceptual markers to each door, at a penalty. These markers must be in the same position as the visual cues. The office will contain real office furniture: chairs, tables, desks, bookcases, file cabinets. There will be enough room for the robot to have a clear path to each of the doors. At some penalty, teams may modify the environment by placing boxes or cardboard strips to fill out convex forms for objects so that their robots will be able to sense them better (normal sonar placement makes them particularly bad at detecting table tops, for example, leading to a danger of decapitation). We will use typical office furniture: chairs with single center posts, chair with four skinny legs, tables with four skinny legs, desks with drawers, etc. Experiment with your own office furniture. The robot is started at an arbitrary location in the office, facing the finish line. The robot will have completed the course when it comes to a complete stop within 2 meters of the finish wall. Teams may add, at some penalty, a maximum of two perceptual markers to the finish wall. These markers will be placed on two designated panels of the wall. Multibots will all be started in the office at some arbitrary location, facing the finish line. We will try to make this a side-by-side competition, with two identical arenas placed next to each other, and two robots running at the same time. Depending on the success of the robots, we may have a final single-elimination round. Event II: Blocks-world stacking A typical configuration is in the Postscript file event2.ps. The arena area will be 15 meters by 15 meters. A partition will intrude into the arena, of the approximate size and position given in the diagram. A set of pushable blocks and immovable obstacles will be scattered in an open area. Robots are to push the blocks into a pattern specified at the start of the event. The pattern will contain from 4 to 6 normal and special blocks. Points are awarded for speed and accuracy. Maximum time limit is 30 minutes. All blocks should slide rather than tip over when pushed from near the bottom. Blocks will be light enough to be pushed by Nomad-type robots; we will try to keep floor friction to a minimum. There have been some requests by teams to modify blocks so that they can be pulled or manipulated in some way. Since the blocks are closed on all sides, this would mean attaching hooks or other protrusions. We will allow teams to do this, as long as they are not used as special perceptual markers. To enforce this, we will ask teams to also place their protrusions on the obstacles. The protrusions should not harm the boxes (taped-on would probably be ok). There are three types of blocks: obstacles, normal blocks, and special blocks. All will be marked with visual cues on one or two sides, and will have shape cues. Normal blocks will have crosses as in the first event, and special blocks will have X's: crosses rotated 45 degrees. The obstacles are marked with a square O about 12 inches in diameter, (see the Postscript marks.ps file). Markers will be placed near the top and near the bottom of each block. Shape cues are that obstacles are rectangular, with one side at least 20cm longer than the other; normal blocks are square and larger than 80cm on a side; special blocks are square and smaller than 80cm. At some penalty, teams may attach their own markers to the sides of objects, at most two markers per object, at the same place as the visual cues. A pattern to be formed will be given just before the start of the event. This pattern will consist of a set of four to six blocks pushed together near the free end of the center partition (see the diagram). Extra points will be awarded for having special blocks at the places noted in the pattern; others positions may be filled by either normal or special blocks. The diagram has an example pattern. Blocks in the pattern need not touch, but they should be close. There will be at least enough blocks to form the pattern; typically there will be more of each kind than necessary. Robots are allowed to push any block except the obstacles; however they should not harm the blocks by running into them at high speed. Robots should not try to push the walls. The initial placement of blocks will be random. Some blocks will be aggregated into groups, and some will be alone. Aggregates will not contain obstacles, so robots can push these apart without fear of moving obstacles. There are no guarantees that perceptual markers will be visible for all boxes in an aggregate. There are no markers on any of the walls. At some penalty, teams may place at most 2 markers at locations of their choosing on the walls. Multibots will be started at or near the same location and in the same orientation as a single robot. Event III: Office delivery A typical configuration is in the Postscript file event3.ps. The arena area will be approximately 15 meters by 23 meters. The arena will be divided by walls into offices and open areas, which include corridors. All corridors will be at least 2 meters wide. The arena and offices will be approximately rectilinear, but there may be some slanting segments. A short time before the event, teams will be given a map with an approximate description of the actual arena. This map will look like that in event3.ps; it will also have a meter grid on it, so that approximate dimensions can be read off. The task is to find a marked object placed somewhere in the arena, and bring it to a specified location. No manipulation is required, the robot must just approach the object indicate it has found the object to "acquire" it. For bonus points, the robot may actually pick up the object and carry it. The delivery object is a glass coffeepot with a handle. It will be located in one of the offices, on top of a block. The block will be marked on each side with the cross from Event I. As a shape cue, the block will be less than 80cm on each side. All other blocks inside offices will be larger than 80cm on each side. The robot will be started in an unknown position and orientation within the arena. A hint will be given about its position: whether it is in an office or a corridor, and which quadrant of the arena it is in. Another hint will give the quadrant in which the delivery object is to be found. The robot must navigate to the object, declare it has found it, and "deliver" it to a marked office specified at the start. The corridors and doorways will be wide enough to accommodate the widest robot. There will be obstacles, and they may block some corridors and doorways; but there will always be an alternative way into all offices. Wider robots might be at a disadvantage here. Obstacles are blocks or groups of blocks. They will not have any perceptual markers on them, except for the delivery object block. Some doorways may be "closed", so that wall partitions cover them and they are blocked. The actual arena map will be handed out on the day of the event. It will show the locations of all walls, but no obstacles or the delivery object. It will also show the locations of most of the office doors. Some doors may not be indicated on the map, and some may be "closed" as indicated above. The map will be reasonably accurate; the error in distance between any two features should be less than 5%. About half of the office openings will be perceptually marked with crosses ("+" marks) on one side or the other of the opening. There will be two of these markers, one near the top and one near the bottom. Additionally, there will be a unique identifier code next to each cross, consisting of a set of vertical or horizontal stripes each about 2 inches wide (see the marks.ps file). The map will indicate which openings have the markers, and what their identifier is. Teams may place their own perceptual markers, at some penalty, in the same location as the crosses. There have also been some inquiries about using more realistic visual cues for the office codes. More realistic cues include small-size alphanumerics, similar to office door labels in current buildings. Teams will be allowed to replace the codes with their own cues at no penalty, as long as they are passive, small, and human-readable. The replacements must go in the same position as the original identifiers. Multibots will be started in two locations, and a separate hint given for each location. Points are awarded for speed, safety and partial completion of the task. Maximum time is 30 minutes.