Robots – Cooperative Navigation Systems
Mr. Clean Reports for Service
Self-propelled autonomous service robots have been cleaning the floors of several supermarkets with a high degree of reliability for the last three years. The robots are piloted by SINAS, the world's most advanced robotic navigation system. At Siemens, researchers are now working on systems that will allow these machines to work together as teams.
Ever reliable, autonomous cleaning machines produce smiles as well as sparkling clean floors
The vacuum cleaner robots that have recently entered the market must feel as lost as a captain without a compass. Tirelessly focused on cleaning carpets, they wander about aimlessly in private households. After about an hour's work, they head for their recharging stations; then, after a short interlude, these fully automatic helpers are set for action again. Unfortunately, they don't have a clue about which areas they have already cleaned and which ones still need to be cleaned.
Attention, Shoppers. But not all robots are created equal. When it comes to the ST82 R, which is guided by a SINAS navigation system, things are very different. Shoppers at the Albert Heijn B.V. chain of supermarkets in the Netherlands are more than a little surprised when a big floorbot asks them politely to step aside, saying something like: "Excuse me, but I'd like to clean the floor here." The robot has been navigating the supermarket's aisles on its own, polishing the floors to a gleaming sparkle since the fall of 1999.
The ST82 R was developed jointly by Hefter Cleantech and Siemens. "It has taken years of development work to perfect the intelligent pilot," says Project Manager Gisbert Lawitzky, a member of the Intelligent Autonomous Systems Department at Siemens Corporate Technology in Munich, Germany. "The idea and initial prototypes of SINAS, which stands for Siemens Navigation System for Autonomous Service Robots, originated at our research laboratories in Munich. SINAS was then developed to the product stage in collaboration with the Siemens Automation and Drives Group. Today, our system is the world's most advanced navigation aid for self-propelled machines," says Lawitzky.
SINAS is useful not only for cleaning, but also for all types of transportation tasks. To navigate successfully, the pilot system must always know where it is and if there are any obstacles to avoid. SINAS achieves this by using sensors that supply it with continuous information regarding its position and immediate environment.
Seeing and Being Seen. Laser scanners and ultrasound systems serve as ST82 R's eyes. "They enable cleaning robots to reliably detect persons or shopping carts in their paths, and to slow down or stop in time," explains Lawitzky. If there is no response to a request to step aside, the robot skillfully maneuvers around the obstacle and then returns to its route. Sensor signals are processed in the machine's brain, a controller that performs complex computational processes. "The software we've developed evaluates all incoming environmental information 20 times per second and issues situation-specific commands to the machine's steering mechanism in realtime," reports Lawitzky. "Aside from determining its position and reacting to possible problem situations, the robot of course occupies itself with its primary task of scrubbing and drying the floor. Furthermore, I should add that its secondary job is to attract attention as an advertising medium: As it moves among shoppers, the robot announces advertising messages, plays music and displays advertisements."
Before it can be released for independent operation, the service robot must get to know its working environment. This learning process is directed by a human coach, who initially guides the robot through the area it will be expected to clean.
"What's so special about SINAS is the fact that it doesn't need any external navigational aids, such as reflective strips or guide wires," says Lawitzky. "The robot collects the initial location data through special odometers mounted on its axles and with the aid of a gyrocompass, which provides continuous estimates of its orientation." Since these data are too imprecise, an apprentice robot continuously records additional information about its position by scanning its environment and memorizing readily recognizable structures such as sections of walls, shelves and columns. Once this phase has been completed, the machine automatically combines such landmarks in a map that aids it in its autonomous locomotion.
During a second guided tour, the coach explains which areas the robot is expected to clean and which are off-limits. It never even gets to see certain areas.
During autonomous operation, the robot follows exactly the same paths it was shown initially—provided, of course, it doesn't encounter obstacles. During its progress, it continuously compares the route it has memorized with with its current route by visualizing its environment and comparing it with a stored map.
This mobile experimental robot scans its surroundings by using eight eyes that conceal ultrasound sensors. A sixth eye is a laser scanner (black box with the greenish light). The box above it contains a gyrocompass. A controller (gray box at rear) and SINAS handle navigation
New Worlds. Smart cleaning robots are currently employed mainly in supermarkets. But they're entirely capable of exploring other new worlds, such as airports, railroad stations, factories, trade fair centers and hospitals. According to a study by the International Federation of Robotics, more than 3,000 cleaning robots will be carrying out tasks in non-domestic applications by 2005 (see Robots: Facts and Forecasts). But cleaning an entire airport terminal would be expecting a bit much of a single robot. Such a job would have to be shared by a team—and that means that each member would have to know exactly what tasks it was expected to perform. Just as with a human team, the work would progress as efficiently as possible. The good news is that the sort of problems that tend to complicate or disrupt teamwork among humans don't seem to undermine tcooperation among robots. That's the conclusion of research scientists at the Department of Intelligent Autonomous Systems at Siemens Corporate Technology in Munich, Germany.
Simulations show a team of robots cleaning a supermarket in Bussum (The Netherlands). The area is divided into cells of similar size. When the communications radii of two robots intersect (right picture), the robots communicate which cells have already been cleaned and which ones still need to be tackled (white area). The strategy is designed to avoid redundant work
Pilot projects with three mobile test robots named R, G and B have demonstrated that the robots can perform cleaning jobs as a team while skillfully avoiding collisions. This ballet is choreographed and directed by ABCHOR, which stands for Agent Based Control Architecture for Heterogeneous Open Robot Communities. "The basic idea is that each robot is controlled by a number of software agents, and each of these agents is assigned to a particular task.
When it comes to navigation issues, for instance, there is a SINAS-based Navigation Agent," explains Markus Jäger, an information scientist. "There is also a GoalAgent that tells the NavigationAgent where it must go. A distinguishing feature of ABCHOR is that a given robot's agents not only communicate with its other onboard agents, but that they can also use radio communications to make contact with the agents of other robots" (for more information about technology involving agents, see articles on Agents, Bots And Avatars in Pictures of the Future, Fall 2001).
Team Spirit. There are two basic methods of dividing tasks among robots—static or dynamic. "In a static division of the work area, machines clean only a predefined section. Here, the drawback is that the system can't adapt to new situations," explains Jäger. He adds that if a team member should fail or require more than the expected amount of time for its sector, some of the work simply wouldn't get done, or cleaning the entire area would take too long.
The three experimental robots have therefore been programmed for dynamic operation. They don't do any actual cleaning just yet, since the present phase of the program is focused on demonstrating "team spirit." But eventually their brains will be implanted in real cleaning robots.
If an airport robot, for instance, is forced to pause because the line at the check-in counter is long and passengers are blocking its path, the machine attempts to contact its colleagues by radio. Its signals have a range of about 10 m and are communicated not by a global network but by a wireless LAN (see article UMTS and More in Pictures of the Future, Spring 2002). Depending on the situation, other service robots can arrive promptly to help clean the area as soon as it's clear.
Cooperative Behaviors. The first step to ensure smooth cooperation is to divide the area to be cleaned into cells of approximately equal size and to assign a number to each cell. Since all of the team members carry the same map in their heads, they can understand the meaning of information such as "Hello, I am currently cleaning the floor in cell number five." In other words, the other robots know that they should stay out of cell number five.
"At this point," says Jäger, "there are basically two scenarios. If all of the robots are located far away from one another in different starting positions, it's not necessary that they coordinate which team member works which area. Each member assumes it will clean the nearest cell, and simply goes about its work." Only when they happen to encounter each other during their work—in other words, when their 10-m communications radii intersect, do they tell each other which areas have already been cleaned by whom to avoid redundant work.
"All previous approaches have been based on the assumption that the machines must be in constant communication with each other," reports Jäger. "However, we are using an entirely new approach. Our current research shows that robots can intelligently divide the area to be cleaned among team members without constant communication with each other."
In the second scenario, all robots start out from the same location. "If for some reason they have to begin their cleaning job in the same cell, it's a must that they communicate and coordinate with each other," says Jäger. But they don't need much discussion to agree—their decision can be instantaneous. It's reached by applying random algorithms. "A given section could, for instance, be assigned to the robot with the highest ID number. On the other hand, the robots could perform a virtual dice game in which the one with the highest score is assigned the section," says Jäger.
But regardless of the type of radio message, the objective is the same: to ensure that several robots don't clean the same area simultaneously. If two units approach each other too closely in their work, they exchange information regarding their current locations and their intended routes. If a collision hazard ensues, one unit pauses and yields the right of way to the other. Exactly which robot does what depends on predefined criteria. What matters is that the best decision is reached for the system as a whole. As a case in point, it wouldn't make sense to stop the faster-moving of the two robots. So the slower-moving unit yields the right of way until the collision hazard has passed.
"We are currently working not only to improve the way multiple robots share the work area, but also to make their route planning as efficient as possible," reports Jäger. "We've performed a great many computer simulations. Now we're conducting extensive tests in a real environment." As far as the future is concerned, this means that cleaning robots may not have to work alone for much longer. Instead, they can look forward to being part of a team!
Ulrike Zechbauer