16. Filtering and State EstimationΒΆ

Consider the two robots given in Fig. 16.1. How should we approach designing robots to function in conjunction with humans or instead of humans? For example:

  • Robotic system to move goods through a distribution center.
  • Robotic system to care for the elderly.
  • Robotic system to perform tasks in hostile environments (deep sea, reactors, space, underwater caves, etc).
  • Robotic systems for assistive technology.

To do this we need to have perception of a changing environment, we need to make decisions about how to respond based on the robot function (goals) and we need to control effectors to carry out the intended functions. For robot perception alone, we scan the environment, segment out objects and then recognize the objects. There many types of sensors used to just understand the surrounding environment:

  • Contact sensors
  • Internal Sensors
  • Accelerometers
  • Gyroscopes
  • Compasses
  • Proximity Sensors
  • Sonar
  • Radar
  • Laser range finders
  • Infrared
  • Cameras
  • GPS

One of the basic functions, localization, is completely dependent on the sensors. Decisions about future actions are made based on the sensors. Feedback from the actuators is also based on the sensors. Many aspects of the system ride on the sensors. The Sensors Chapter presented a number and variety of sensing systems. There is a vast array of sensors which can sense or measure physical quantities. Accuracy on sensors varies greatly with some very accurate and others having considerable errors.


Fig. 16.1 Examples of two robotics systems that interact with humans.


Problem: How can we design a system which can function without human input but has random elements in the environment? How can the system determine its location accurately enough to navigate? How can the robot use manipulators around humans safely if manipulator location, feedback and control are uncertain. In manufacturing systems, we are able to instrument objects of interest and highly constrain the environment. The robot might not be local and the object to be manipulated may have a known location. Clearly variation and noise occur, but not to the degree found in robots that are intended to go into the world and operate outside the confines of a factory.