Potential field methodologies

In potential field methodologies, behaviors 20211217202757 are described as vectors and use vector summation to combine the output of different behaviors. For example, the output vector of a “move-forward” and a “avoid obstacle” would be combined, defining the movement vector.

Potential field

Potential fields area arrays that represent the world on an (x,y) grid, and on every cell of the world there is a vector that represents a force, like in a magnetic or gravitation field.

There are 5 types of primitive fields:

  • Uniform: same everywhere
  • Perpendicular: perpendicular to object/wall
  • Attraction: vectors point to object
  • Repulsion: vectors point away of object
  • Tangential: vectors spin around object

Magnitudes of the vectors are defined using magnitude profiles. These magnitudes can be describe on many different ways, such as constant, linear, exponential…

Programming of potential fields can be easily done bu thinking of each field as a single function.

Behavior combination

As stated before, the combination of behaviors is done by perfomring a summation of the vectors. However, some key points have to be taken in consideration:

  • The update rate of the robot will determine if the robot overshoots or not the goal, as the magnitude changes the velocity vector.
  • The ability to perform certain movements for the robot has to be taken in consideration. Instantaneous changes on direction/velocity may not be possible.
  • There can be some points where the sum of the vectors produce a 0.0 vector, producing a local minima where the robot cannot move from. Solutions for these are: use small random vectors; implement the “avoid” behavior as an intelligent behavior that receives all other vectors as input (Navigation-Templates); harmonic functions (expensive)

Notes References

20210711201545 Robotics Basics - Reactive Paradigm

20211217202757 Robot behaviors

20210514183815 INDEX - Robotics

References

(Murphy 2000)

Murphy, Robin. 2000. Introduction to AI Robotics. Intelligent Robotics and Autonomous Agents. Cambridge, Mass: MIT Press.