Dilip Kumar Pratihar

A genetic-fuzzy approach for mobile robot navigation among moving obstacles

In this paper, a genetic-fuzzy approach is developed for solving the motion planning problem of a mobile robot in the presence of moving obstacles. The application of combined soft computing techniques - neural network, fuzzy logic, genetic algorithms, tabu search and others - is becoming increasingly popular among various researchers due to their ability to handle imprecision and uncertainties that are often present in many real-world problems. In this study, genetic algorithms are used for tuning the scaling factors of the state variables (keeping the relative spacing of the membership distributions constant) and rule sets of a fuzzy logic controller (FLC) which a robot uses to navigate among moving obstacles. The use of an FLC makes the approach easier to be used in practice. Although there exist many studies involving classical methods and using FLCs they are either computationally extensive or they do not attempt to find optimal controllers. The proposed genetic-fuzzy approach optimizes the travel time of a robot off-line by simultaneously finding an optimal fuzzy rule base and optimal scaling factors of the state variables. A mobile robot cant then use this optimal FLC on-line to navigate in presence of moving obstacles. The results of this study on a number of problem scenarios show that the proposed genetic-fuzzy approach can produce efficient knowledge base of an FLC for controlling the motion of a robot among moving obstacles.

Time-optimal, collision-free navigation of a car-like mobile robot using neuro-fuzzy approaches

Neuro-fuzzy approaches are developed, in the present work, to determine time-optimal, collision-free path of a car-like mobile robot navigating in a dynamic environment. A fuzzy logic controller (FLC) is used to control the robot and the performance of the FLC is improved by using three different neuro-fuzzy (NN-FLC) approaches. The performances of these neuro-fuzzy approaches are compared among themselves and with those of three other approaches, such as default behavior, manually-constructed FLC and potential field method, through computer simulations. The neuro-fuzzy approaches are found to perform better than the other approaches, in most of the test scenarios. Moreover, the performances of both the genetic algorithm (GA)-optimized NN-FLC (Mamdani Approach) as well as GA-optimized NN-FLC (Takagi and Sugeno Approach) are seen to be comparable. It is also interesting to note that the CPU times of all these approaches are found to be low. Thus, they might be suitable for on-line implementations.

Optimal path and gait generations simultaneously of a six-legged robot using a GA-fuzzy approach

This paper describes a new method for generating optimal path and gait simultaneously of a six-legged robot using a combined GA-fuzzy approach. The problem of combined path and gait generations involves three steps, namely determination of vehicle’s trajectory, foothold selection and design of a sequence of leg movements. It is a complicated task and no single traditional approach is found to be successful in handling this problem. Moreover, the traditional approaches do not consider optimization issues, yet they are computationally expensive. Thus, the generated path and gaits may not be optimal in any sense. To solve such problems optimally, there is still a need for the development of an efficient and computationally faster algorithm. In the proposed genetic-fuzzy approach, optimal path and gaits are generated by using fuzzy logic controllers (FLCs) and genetic algorithms (GAs) are used to find optimized FLCs. The optimization is done off-line on a number of training scenarios and optimal FLCs are found. The hexapod can then use these GA-tuned FLCs to navigate in test-case scenarios.

Tuning of neural networks using particle swarm optimization to model MIG welding process

Abstract Particle swarm optimization technique has been used for tuning of neural networks utilized for carrying out both forward and reverse mappings of metal inert gas (MIG) welding process. Four approaches have been developed and their performances are compared to solve the said problems. The first and second approaches deal with tuning of multi-layer feed-forward neural network and radial basis function neural network, respectively. In the third and fourth approaches, a back-propagation algorithm has been used along with particle swarm optimization to tune radial basis function neural network. Moreover, in these two approaches, two different clustering algorithms have been utilized to decide the structure of the network. The performances of hybrid approaches (that is, the third and fourth approaches) are found to be better than that of the other two.

Effects of turning gait parameters on energy consumption and stability of a six-legged walking robot

Minimization of energy consumption plays a key role in the locomotion of a multi-legged robot used for various purposes. Turning gaits are the most general and important factors for omni-directional walking of a six-legged robot. This paper presents an analysis on energy consumption of a six-legged robot during its turning motion over a flat terrain. An energy consumption model is developed for statically stable wave gaits in order to minimize dissipating energy for optimal feet forces distributions. The effects of gait parameters, namely angular velocity, angular stroke and duty factors are studied on energy consumption, as the six-legged robot walks along a circular path of constant radius with wave gait. The variations of average power consumption and energy consumption per unit weight per unit traveled length with the angular velocity and angular stroke are compared for the turning gaits of a robot with four different duty factors. Computer simulations show that wave gait with a low duty factor is more energy-efficient compared to that with a high duty factor at the highest possible angular velocity. A stability analysis based on normalized energy stability margin is performed for turning motion of the robot with four duty factors for different angular strokes.

Forward and reverse mappings in green sand mould system using neural networks

The quality of castings in a green sand mould is influenced significantly by its properties, such as green compression strength, permeability, mould hardness, and others, which depend on input parameters. The relationships of these properties with the input parameters, like sand grain size and shape, binder, water, etc. are complex in nature. In the neural network based forward mapping, mould properties are expressed as the functions of input parameters, whereas attempts can also be made to determine an appropriate set of input parameters, to ensure a set of desired properties, in reverse mapping. In the present work, the problems related to both the forward as well as reverse mappings in green sand mould system were tackled by using a back-propagation neural network (BPNN) and a genetic-neural network (GA-NN). Batch mode of training had been provided to both the networks with the help of one thousand data, generated artificially from the regression equations obtained earlier by the authors. The performances of the developed models had been compared among themselves for 20 randomly generated test cases. The results show that GA-NN outperforms the BPNN and that both the NN approaches are able to carry out the reverse mapping effectively.

Design of a genetic-fuzzy system to predict surface finish and power requirement in grinding

We have developed, in this paper, a genetic-fuzzy system, in which a genetic algorithm (GA) is used to improve the performance of a fuzzy logic controller (FLC). The performance of an FLC depends on its knowledge base (KB), which consists of both the data base (membership function distributions of the variables) as well as rule base. In the developed genetic-fuzzy system, the KB of the FLC is optimized, off-line, using a GA. Three approaches are developed, in the present work. In the first approach, the membership function distributions of the variables are assumed to be triangular, whereas a second-order polynomial function and a third-order polynomial function are used in the second and third approaches, respectively. The results of these approaches are compared for making prediction of surface finish and power requirement in grinding, a machining process used to generate smooth surface on the job. For some of the test cases, comparisons are also made of the results predicted by the genetic-fuzzy system with those obtained through the real experiments.

On-line stable gait generation of a two-legged robot using a genetic–fuzzy system

Abstract Gait generation for legged vehicles has since long been considered as an area of keen interest by the researchers. Soft computing is an emerging technique, whose utility is more stressed, when the problems are ill-defined, difficult to model and exhibit large scale solution spaces. Gait generation for legged vehicles is a complex task. Therefore, soft computing can be applied to solve it. In this work, gait generation problem of a two-legged robot is modeled using a fuzzy logic controller (FLC), whose rule base is optimized offline, using a genetic algorithm (GA). Two different GA-based approaches (to improve the performance of FLC) are developed and their performances are compared to that of manually constructed FLC. Once optimized, the FLCs will be able to generate dynamically stable gait of the biped. As the CPU-time of the algorithm is found to be only 0.002 s in a P-III PC, the algorithm is suitable for on-line (real-time) implementations.

Evolutionary robotics—A review

In evolutionary robotics, a suitable robot control system is developed automatically through evolution due to the interactions between the robot and its environment. It is a complicated task, as the robot and the environment constitute a highly dynamical system. Several methods have been tried by various investigators to solve this problem. This paper provides a survey on some of these important studies carried out in the recent past.

Soft computing-based gait planners for a dynamically balanced biped robot negotiating sloping surfaces

Dynamically balanced gait generation problems of a biped robot moving up and down the sloping surface have been solved utilizing soft computing-based approaches. The gait generation problem of a biped robot is difficult to model due to its inherent complexity, imprecision in the collected data of the environment, which are the characteristics that can be the best modeled using soft computing. Two different approaches, namely genetic-neural (GA-NN) and genetic-fuzzy (GA-FLC) systems have been developed to solve the ascending and descending gait generation problems of a two-legged robot negotiating the sloping surface. Two modules of neural network (NN)/fuzzy logic controller (FLC) have been used to model the gait generation problem of a biped robot using the GA-NN/GA-FLC system. The weights of the NNs in the GA-NN and knowledge bases of the FLCs in the GA-FLC systems are optimized offline, utilizing a genetic algorithm (GA). Once the GA-NN/GA-FLC system is optimized, it will be able to generate the dynamically balanced gaits of the two-legged robot in the optimal sense.