Multisensor data fusion for robotic control

Nowadays robots perform more and more tasks, from simpler ones such as automatic domestic vacuum cleaner, to highly skilled ones such as tele-surgery. To do so a robot continuously needs to know what to do and what has been done. A user controls robots through a system controller, which processes both his commands and information from the environment. These information are collected by a measurement system, often called observer, which bring back them to the system controller. Measures are the information used to perform the control. The simpler measurement system can, in general, be considered as made up of two elements: 1. A sensor or transducer which is an element that produces a signal relating the quantity being measured. Sensors are elements that when subject to some physical change experience a related change. 2. A signal conditioner which takes the signal from the sensor and processes it to make it suitable for the specific application. The signal may be, for example, too small or too noisy. Different kinds of sensors are used in robotics, such as position, velocity, force sensors and so on. Each application needs the right sensor to be chosen, the choice must satisfy different criterions such as performance, cost and feasibility. In this way both static, as range or sensitivity, and dynamic characteristics, as response time, of the sensor have to be considered. In robotic applications there are two main subclasses of sensors: internal and external. The former bring information about the robot; the latter bring information about the environment. In a mobile robot, for example, encoders are internal sensors, while ultrasound sensors are external. Multisensor data fusion seeks to combine data from multiple sensors to perform inferences that may not be possible from a single sensor. Different sensors have different strengths and weaknesses, so fused data from multiple sensors provides several advantages over data from a single sensor: A wider and more accurate range of information by combining data from different types of sensors. Redundancy. Increased reliability. Weaknesses compensation. Sensors failure detection and handling. Typical applications that can benefit from multiple sensors are, first of all, mobile robot navigation, target tracking, aircraft navigation and industrial tasks control. There are a number of different ways to integrate or fuse information provided by multiple sensors. Multisensor input system controller The simplest approach to multisensor data fusion, is to use the information from each sensor as a separate input to the system controller. This approach may be the most appropriate if each sensor is providing information concerning completely different aspects of the environment. The major benefit gained through this approach is the increase in the extent of the environment that is able to be sensed. The only interaction between the sensors is indirect and based on the individual effect each sensor has on the controller and so on the whole robot. If there is some degree of overlap between sensors, concerning some aspect of the environment they are able to sense, it may be possible for a sensor to directly influence the operation of another one. In this way the value of the combined information that the sensors provide, is greater than the sum of the value of the information provided by each sensor separately. This synergistic effect can be achieved either by using the information from one sensor to provide cues, or guide the operation of other sensors, or by actually combining or fusing the information from multiple sensors. The whole research activity has dealt with two main topics: the RAPOLAC Project and the Mobile Robots. The former are discussed in the first four chapters while the latter in the following four.