Avoiding unexpected obstacles is a fundamental requirement for utility in a personal robot. Virtually none of the mobile service oriented personal robots designed, built, and sold to date have this basic capability to the degree needed in a dynamic home environment. Simple room-to-room errand running, efficient vacuuming, security patrolling, etc. all require this ability. For a personal robot to be useful and provide timely, beneficial services to humans, the robot must have enough intelligence to find it's way, for example, to the back bedroom in spite of unanticipated obstacles such as dropped toys, etc. Cognizant navigation is a non-trivial problem that has a number of facets.

There must be enough sensor information of the right kind to not hit typical obstacles such as walls, furniture, and people. There must also be enough sensor information to avoid smaller obstacles such as toys. Furthermore, the brain must be able to react to quick local changes without sacrificing the ability to give the robot tasks to complete. The robot must also have a memory of where it is within the world and be able to repeatedly find locations within that world even if there are unexpected obstacles. This means that there must be enough processing power and RAM to accomplish this while still having enough battery life to stay active for many hours.. These important capabilities are the basic, required foundation for useful robots in a human environment. Until the GSI MSR, almost all mobile service robots have fallen short in one or more of these areas.

Many available robots are limited by their sensors. Mobile robot sensors such as bump switches, feelers, and whiskers have the problem that they cannot sense their environment without physical contact with the world. Fixed single sonar and infrared distance (IR) range finders are an improvement, but they give very little infor-mation about the surrounding world. They may help avoid running into an obstacle directly in front of the robot in one direction, but they aren't very useful in helping the robot navigate.

Even having many different kinds of these sensors does not necessarily solve the problem. The robot must be able to assess the current space around the robot to allow robust navigation, and to do that the robot must have enough information of the right kind, not just many arbitrary sensors. Expensive research robot manufacturers understand this need, and solve this problem through an array of multiple sonar and IR sensors or even more expensive machine vision systems.

The GSI MSR solution to this problem is a high-data, low cost sonar and IR sensor scanner that can image the surrounding space in 75 directions, not just one single direction. Most research robots may have as many as 32 directions that can be monitored, but some have as few as 7!.

The GSI MSR uses a scanning sonar and IR range finding array giving 25 sensor positions on each complete left to right back to left scan every 2-3 seconds. With only one sonar and two IR range finders, we now have 75 "virtual" sensors on the front of the robot. This is extremely high granularity. Managing this much sensor data using traditional if-then solutions is non-trivial and would consume a not insignificant amount of programming time. And still result in an undesirable, brittle solution requiring a highly structured, "clean" environment. GeckoSystems' GeckoBrain system architecture does not use a traditional reactive, bump-turn, solution. We use subsumptive fuzzy logic (See hypertext link.) in our hybrid architecture.

Many robots are limited by their brain architecture. Historically, personal robot architectures have been based on either a pre-set path following technique, where the sensors are only used to detect failure of the preprogrammed path, or they have used a purely reactive technique that has no concept of the larger world that the robot inhabits and cannot be easily applied for useful tasks.

The path-following techniques suffer from being unable to adapt to changing conditions quickly or smoothly. The robot basically travels blind until it is about to hit something, and once it has detected an obstacle, the resulting decisions required are very complex. As a result, the environment must be highly structured to avoid confusing the robot so that simple decisions will suffice or a lot of computing power must be available to maintain and compute path alternatives. Requiring a highly structured environment reduces the usefulness and flexibility of such a robot in a human environment. Requiring a lot of on board processing power makes robots really expensive.

The purely reactive architectures suffer from having little sense of past events, future goals, or even where exactly it is within the world. Typically such robots have no memory of the world that they have traveled and "live" only instant to instant. They may reach a particular destination, but it is by pure chance and the robot will not be able to recognize that it has reached the desired destination without providing a modified environment (e.g. beacon techniques). In its pure form, something seen in many toy robots, this technique is almost useless for useful navigation or tasks in a dynamic human environment.

The GSI MSR's brain is different. Its Fuzzy Hybrid Architecture provides the benefits of both control and reaction within a single framework without the disadvantages of either technique alone. As a result, it is able to respond quickly and intelligently to short term navigation situations while still providing the ability to guide the robot toward accomplishing useful tasks within a map of the world that the robot maintains. It turns out that this approach is synergistic and reduces the complexity of trying to "force fit" either of the other traditional solutions to solve the whole problem.

Many robots are processor weak and battery poor. Most embedded processors simply do not have the necessary RAM or CPU speed to handle the required navigation, sensor filtering, and mapping capabilities for Cognizant Navigation. Furthermore, robots that do have this amount of processing power typically put them on the robot, which can draw 90 watts of vital battery power and severely reduces the length of time that the robot can stay "on" doing useful tasks (typically no more than a couple of hours). This also increases the cost of the robot substantially.

An important attribute of a robot capable of cognizant navigation is the ability to compensate for sensor errors and environmental discontinuities and hazards. This implies a memory within the robot's control system such that the robot knows where it has been, where it is, and where it is ultimately to arrive. This is an endpoint to seek, or room to find.

In other words, if the robot can't "remember" to take advantage of the information, then it shouldn't be considered usefully aware or cognizant. An example would be a simple contact switch that stops the robot if it hits a wall. If the robot just sits there and doesn't use the new information that a wall is there to help it complete the course, then it would not be considered under cognizant navigation. Certainly, a robot that can only sense when physically contacting an obstacle is no more cognizant that a toy car that bumps into a wall, reverses momentarily, and then proceeds at a new angle to its initial encounter with the obstacle. Hence, if the robot recovers the path and reaches the initial goal, even with the new, added obstacles, then it is showing cog-nizant abilities.

This is a behavior the GSI MSR can demonstrate "out-of-the-box" since it has a virtually unlimited memory and data storage in the PC you already have. The underutilized PC, can now be used for useful tasks in house cleaning, security, and child or eldercare, when you would not be normally using your PC. (The GSI MSR avoids all these problems by making use of your powerful home PC that you already have via a wireless link. In stark contrast, the GSI MSR can easily stay "awake" and patrol an area for an incredible 20 plus hours at a time. Vacuuming can reduce this, but even here 6 to 8 hours of "on" time can be expected.)

As we have discussed, cognizant navigation is much more than simple reactive, bump-turn mobile robot behavior. It may reach the goal, but isn't "aware" that it is attempting to reach the goal at some level. Line followers may be "aware" that they are trying to reach a goal, but they have problems when reacting to new situations due to the brittleness of their complex if-then programming. Now you see the additional layer of complexity. The word "aware" implies the robot remembering where it is, where it was, and where it is "supposed" to be going. Cognizant means to be aware or have conscious knowledge. Humanlike short term and long memory management (desired event retention) is the key. This requires the power of the Personal Computer. Your existing PC has the raw computing power, memory, and data storage needed for robust personal robot cognizant navigation, scheduling of areas to be vacuumed, and much, much more.

This goal awareness and ability to return to seeking a predetermined goal is one important attribute that separates the toybots, and most, if not all, hobbybots, from the $30,000 to $100,000 multi-Pentium researchbots. That is until GeckoSystems changed the rules with its leading edge GeckoBrain software. No other commercially available personal computer based, utilitarian robot has ever achieved this high level, extremely difficult, degree of cognizant navigation. The Fuzzy Hybrid Architecture of the Gecko-Brain makes this possible.

Given the foregoing discussion, the reader may now have a better understanding of the importance of a personal robot as a natural peripheral to an existing PC and the useful benefits, such as truely cognizant navigation, that may be realized.

1. Intention to reach a goal and know when it is reached.
   
2. Ability to alter strategy in the face of changing circumstances.
   
3. The power of a personal computer to cost effectively provide the raw computing power to enable cognizant navigation.

Conclusion:

This is but a brief discussion of some the prerequisite traits necessary (here cognizant navigation) for a personal robot to be useful. Even with voice command and control, telling the robot to go to the children's bedroom from the kitchen is not useful unless the robot can avoid unexpected obstacles such as an errant tennis shoe, and resume its journey to and successfully reach the bedroom. The same holds true for vacuuming unattended. The robot must be capable of using an efficient, time saving pattern to vacuum the high traffic areas of your home or workplace while routinely avoiding unexpected obstacles.

Welcome to the 21st century of mobile service robotics: the GSI MSR.