Artificial Potential Field

The goal generates an attractive force and obstacles generate repulsive forces; the robot moves along the combined force in real time.

The artificial potential field method was introduced by Khatib in papers at ICRA 1985 and IJRR 1986 for real-time obstacle avoidance in robot arms and mobile robots, and was first implemented on a PUMA 560 arm. It treats space as a potential-energy map: potential is lowest at the goal, which produces an attractive force, and it's high near obstacles, which produces a repulsive force (usually only within a limited range). At every control cycle, the robot computes the direction of steepest potential decrease (the negative gradient, i.e. the combined-force direction) and takes a small step that way. Its advantages are very low computational cost, the ability to update in real time from sensor data, and the ability to handle moving obstacles; its main drawback is getting stuck in local minima — when the attractive and repulsive forces happen to cancel out, the robot stalls or oscillates in place — and it doesn't guarantee the shortest path. It's therefore mostly used as a local obstacle-avoidance layer alongside global planners like A* or RRT; later reactive methods such as Riemannian motion policies follow a similar idea.

ExampleA mobile robot needs to pass through a narrow doorway, with a repulsive field on each side of the frame and an attractive field from the goal beyond it; if the doorway is narrow enough, the two side repulsive forces combined can cancel out the attraction, and the robot stalls in front of the door — a local minimum.

Also called

APF, Potential Field Method

Related

Obstacle Avoidance · Path Planning · Rapidly-exploring Random Tree · A* Search · Riemannian Motion Policies · Control Barrier Function

Sources

Khatib, Real-Time Obstacle Avoidance for Manipulators and Mobile Robots (IJRR 1986)
Wikipedia: Motion planning（Artificial potential fields 一节） (Chinese)