Page 69 - Read Online
P. 69

Li et al. Intell Robot 2021;1(1):58-83  I http://dx.doi.org/10.20517/ir.2021.08      Page 64































               Figure 4. Path planning of a mobile robot to avoid local minima with concave obstacles. A: the robot path; B: the landscape of neural
               activity  [13] .

               free navigation and the cooperation of the multi-robot systems. In addition, many developed model variants
               are also discussed for robot path planning.

               3.1.1. Navigation
               The first bio-inspired neural network framework was proposed by Yang and Meng for the mobile robot path
               planning [29] . Many remarkable achievements in mobile robot path planning have been achieved [13,30,31] . Due
               to the global effects of positive neural activity from the target, the robot is not trapped in the undesired local
               minima. Figure 4 shows an example of path generation of a mobile robot to avoid local minima. The robot is
               not trapped in a set of concave obstacles and move to the target position.

               Some researchers consider the different types of robots in the application to navigation. A nonholonomic car-
               like robot was studied by Yang et al. [15,32,33]  for real-time collision-free path planning. The simulation results
               showed the car-like robots performed well in parallel parking, navigation in several deadlock situations, and
               sudden environmental changes conditions. In a house-like environment as shown in Figure 5A, the robot
               moved to the target along the shortest path in case that the door is opened. When the door is closed, the
               robot travels a much longer path to reach the target without any learning procedures. The robot is capable of
               reaching the target along the shortest path without any collisions, without violating the kinematic constraint,
               and without being trapped in deadlock situations.


               In addition, Yang and Meng developed the bio-inspired neural network for robot manipulators [13] . The joint
               space of the robot manipulators was corresponded to the bio-inspired neural network, in which neurons were
               characterized by the shunting model or the additive model. Figure 5B shows the trajectory of robot manipu-
               lators avoiding obstacles. In addition, a virtual assembly system was proposed by Yuan and Yang for assisting
               product engineers to simulate the assembly-related manufacturing process [34] .


               An improved bio-inspired neural network based on scaling terrain was proposed by Luo et al. [35] for reducing
               the calculation complexity. This multi-scale method mentioned better performance in terms of time complex-
               ity. However, the simulation experiments do not give the criteria for choosing the parameter of coarse-scale
               and fine-scale maps. Ni et al. [36]  used a bio-inspired neurodynamics model as the reward function for the
   64   65   66   67   68   69   70   71   72   73   74