Why Introducing Probabilities In Robotics

Uncertainty happens during observation in the real world. Robotics observe the world using sensors, and the uncertainty of sensors is modelled by probabilistic models, such as gaussian distribution.

Uncertainty in Robotics

• What does Robotics do?
  Robotics is a science about controlling mechanical devices (automatically/intelligently) to perceive and manipulate the physical environment.

Word	Description
control	also referred as execute or actuate, which is the function that makes robots interact with the environment.
perceive	also referred as percept or observe, which is the function to transfer environment information to the computing unit of a robot.
manipulate	which means change the environment into desired state.
environment	also referred as world, which can be variable or fixed, including the robot itself and environment independent with the robot

Intelligence of robotic shall be mentioned in another posts. we only talk about uncertainty in this post. Taking a assembly line as an example, the location of the robot is stationary and environment is structured, the operation of robot arms is qualified as long as the operational accuracy is within manageable range. However, the new robot systems operates in a increasingly unstructured environments that are inherently unpredictable, which makes sensors very important in the process of perception.

• What is Uncertainty and why it exist?

  Uncertainty happens in mobile robotics in the following ways:

  • Environments: compared with assembly line, environments of roads are highly dynamic and unpredictable.
  • Sensors: sensors has limited resolution and measurement contains noise.
  • Robots: control noise and wear-and-tear.
  • Models: models are abstraction of the real world, and abstracted model and not model the world fully, which means the robotics system are crude.
  • Computation: due to the requirement of being a real-time system, algorithms usually are approximate to achieve timely response.

Probabilistic Model

• How to model the uncertainty?

The key idea of probabilistic robotics is to represent uncertainty explicitly, using the calculus of probability theory. Instead of find the best guess of what happens in the real world, the probabilistic model represent the information by probability distribution over the space of possible hypothesis. With this model, all sources of uncertainty can be fused mathematically, detailed information shall be mentioned later in later posts, such as kalman filtering.

The most important mindset of applying probabilistic model is that the scalars describing the state of the environment is extended to probabilistic space, instead of being a number.

References

Source: Probabilistic Robotics