Page 72 - Read Online

Page 67 Li et al. Intell Robot 2021;1(1):58-83 I http://dx.doi.org/10.20517/ir.2021.08

Figure 6. Examples of hunting tasks. A: multiple evaders need to be hunted; B: some robots break down [56] .

where is a positive constant and is a monotonically increasing function of the difference between the cur-
rent to next robot moving directions. Compared with path planning and CCPP problems, the main difference
is that many target positions might attract the cleaning robot because all unclean areas are set as targets. Thus,
the turning numbers of clean robots might increase significantly. Function is designed to reduce the turn-
ing numbers. If the robot goes straight, =1; if goes backward, = 0. Thus, the cleaning robot tends to go
straight.

Inthissection,basedonthepreviousknowledgeoftheenvironment,theresearchfieldsofcleaningrobotsusing
the neurodynamics model are categorized as: completed known environment and unknown environment.

3.2.1. Completed known environment
Figure 7 shows the neurodynamics-based CCPP in a completely known environment. The neurodynamics
model can work efficiently in the dynamic environment, so even considering sudden change environment and
moving obstacles in the environment, the cleaning robots can still work efficiently [57,59,60] . In order to improve
the computational complexity, a discrete bio-inspired neural network was proposed to convert to the shunting
equation a difference equation [62] .

One CCPP challenging problem is the deadlock situation. The deadlock area is a specific situation that the
cleaning robot is trapped in a position where all of the neighborhood areas have been covered, but the work en-
vironment is still unclear. If the cleaning robot moves to deadlock areas, the cleaning robot is unable to escape
from the deadlock areas without any interventions. A dynamic neural neighborhood analysis for deadlock
avoidance was proposed based on the characteristics of deadlock areas [59] . The robot can recognize whether
the current position is the deadlock point. If the current position is a deadlock point, the connection weights
of the neural network were changed to generate a path to escape this deadlock point.

3.2.2. Unknown environment
In order to deal with CCPP in the unknown environment, the cleaning robots are typically required to build
a surrounding map with a very limited time range [63] . The onboard sensors have been widely used for robot