|
| Approach | Major Implementation | Strengths | Challenges |
|
| Neural Network | Simulation and real Experiment | (i) Good at providing learning and generalization [232]
(ii) Can help tune the rule base of fuzzy logic [232] | (i) Efficiency of neural controllers deteriorates as the number of layers increase [232]
(ii) Difficult to acquire its required large dataset of the environment during training to achieve best results.
(iii) Using BP easily results in local minima problems [238]
(iv) It has a slow convergence speed [232] |
|
| Genetic Algorithm | Simulation | (i) Good at solving problems that are difficult to deal with using conventional algorithms and it can combine well with other algorithms
(ii) It has good optimization ability | (i) Difficult to scale well with complex situations
(ii) Can lead to convergence at local minima and oscillations [239, 240]
(iii) Difficult to work on dynamic data sets and difficult to achieve results due to its complex principle [233] |
|
| Ant Colony | Simulation | (i) Good at obtaining optimal result [233]
(ii) Fast convergence [234] | (i) Difficult to determine its parameters which affects obtaining quick convergence [240]
(ii) Require a lot of computing time [233] |
|
| Particle Swarm Optimization | Simulation | (i) Faster than fuzzy logic in terms of convergence [132]
(ii) Simple and does not require much computing time hence, effective to implement with optimization problems [192–195]
(iii) Performs well on varied application problems [193] | (i) Difficult to deal with trapping into local minima under complex map [194, 240]
(ii) Difficult to generalize its performance since undertaken experiments relied on objects in polygon form [240] |
|
| Fuzzy Logic | Simulation and Experiments with real robot | (i) Ability to make inferences in uncertain scenarios [129]
(ii) Ability to imitate the control logic of human [129]
(ii) It does well when combine with other algorithms | (i) Difficult to build rule base to deal with unstructured environment [145, 232] |
|
| Artificial Bee Colony | Simulation | (i) It does not require much computational time and it produces good results [241]
(ii) It has simple algorithm and easy to implement to solve optimization problems [236, 241]
(iii) Can combine well with other optimization algorithms [236, 237]
(iv) It uses fewer control parameters [236] | (i) It has low convergence performance [241] |
|
Memetic and Bacterial Memetic algorithm
| Simulation | (i) Compared to conventional algorithms, it produces faster convergence and good solution with the benefit from different search methods it blends [235] | (i) It can result in premature convergence [235] |
|
| Others (Simulated Annealing and Human Navigation strategy) | Simulation | (i) Simulated Annealing is good at approximating the global optimum [242]
(ii) Takes advantage of human navigation that does not depend on explicit planning but occur online [223] | (i) SA algorithm is slow [242]
(ii) Difficult in choosing the initial position for SA [242] |
|
