Soohwan Hyun

Genetic Programming-Based Automatic Gait Generation in Joint Space for a Quadruped Robot

This paper introduces a new approach to developing a fast gait for a quadruped robot using genetic programming (GP). Planning gaits for legged robots is a challenging task that requires optimizing parameters in a highly irregular and multi-dimensional space. Several recent approaches have focused on using genetic algorithms (GAs) to generate gaits automatically and have shown significant improvement over previous gait optimization results. Most current GA-based approaches optimize only a small, pre-selected set of parameters, but it is difficult to decide which parameters should be included in the optimization to get the best results. Moreover, the number of pre-selected parameters is at least 10, so it can be relatively difficult to optimize them, given their high degree of interdependence. To overcome these problems of the typical GA-based approach, we have proposed a seemingly more efficient approach that optimizes joint trajectories instead of locus-related parameters in Cartesian space, using GP. Our GP-based method has obtained much-improved results over the GA-based approaches tested in experiments on the Sony AIBO ERS-7 in the Webots environment. The elite archive mechanism is introduced to combat the premature convergence problems in GP and has shown better results than a traditional multi-population approach.

Genetic programming based automatic gait generation for quadruped robots

This paper introduces a new approach to develop fast gait for quadruped robot using genetic programming (GP). Several recent approaches have focused on using genetic algorithm (GA) to generate gait automatically and shown significant improvements over previous results. Most of current GA based approaches use pre-selected parameters, but it is difficult to select the appropriate parameters for the optimization of gait. To overcome these problems of GA based approach, we proposed an efficient approach which optimizes joint angle trajectories using genetic programming. Our GP based method has obtained much better results than GA based approaches for experiments of Sony AIBO ers-7 in Webots environment. The elite archive mechanism(EAM) was introduced to prevent premature convergence problems in GP and has shown improvements.

An Edge-Set Representation Based on a Spanning Tree for Searching Cut Space

The encoding or representation scheme in evolutionary algorithms is very important because it can greatly affect their performance. Most previous evolutionary algorithms for solving graph problems have traditionally used a vertex-based encoding in which each gene corresponds to a vertex. In this paper, addressing the well-known maximum cut problem, we introduce an edge-set encoding based on the spanning tree - a kind of edge-based encoding. In our encoding scheme, each gene corresponds to an edge subset derived from a spanning tree. In contrast to a traditional edge-based encoding in which each gene corresponds to only one edge, our encoding scheme has the advantage of representing only feasible solutions, so there is no need to apply a repair step. We present a genetic algorithm based on this new encoding. We have conducted various experiments on a large set of test graphs including commonly used benchmark graphs and have obtained performance improvement on sparse graphs, which frequently appear in real-world applications such as social networks and systems biology, in comparison with a scheme using a vertex-based encoding.

A spanning tree-based encoding of the MAX CUT problem for evolutionary search

Most of previous genetic algorithms for solving graph problems have used vertex-based encoding. In this paper, we introduce spanning tree-based encoding instead of vertex-based encoding for the well-known MAX CUT problem. We propose a new genetic algorithm based on this new type of encoding. We conducted experiments on benchmark graphs and could obtain performance improvement on sparse graphs, which appear in real-world applications such as social networks and systems biology, when the proposed methods are compared with ones using vertex-based encoding.

A comparative study between genetic algorithm and genetic programming based gait generation methods for quadruped robots

Planning gaits for legged robots is a challenging task that requires optimizing parameters in a highly irregular and multidimensional space. Two gait generation methods using GA (Genetic Algorithm), GP (genetic programming) are compared to develop fast locomotion for a quadruped robot. GA-based approaches seek to optimize a pre-selected set of parameters which include locus of paw and stance parameters of initial position. A GP-based technique is an effective way to generate a few joint trajectories instead of the locus of paw positions and many stance parameters. Optimizations for two proposed methods are executed and analyzed using a Webots simulation of the quadruped robot built by Bioloid. Furthermore, simulation results for the two proposed methods are tested in a real quadruped robot, and the performance and motion features of GA-, GP -based methods are compared.

Genetic programming-based model output statistics for short-range temperature prediction

This paper introduces GP (Genetic Programming) based robust compensation technique for temperature prediction in short-range. MOS (Model Output Statistics) is a statistical technique that corrects the systematic errors of the model. Development of an efficient MOS is very important, but most of MOS are based on the idea of relating model forecasts to observations through a linear regression. Therefore it is hard to manage complex and irregular natures of the prediction. In order to solve the problem, a nonlinear and symbolic regression method using GP is suggested as the first attempt. The purpose of this study is to evaluate the accuracy of the estimation by GP based nonlinear MOS for the 3 days temperatures for Korean regions. This method is then compared to the UM model and shows superior results. The training period of summer in 2007-2009 is used, and the data of 2010 summer is adopted for verification.

Locomotion Control of 4 Legged Robot Using HyperNEAT

The walking mobility with stability of 4 legged robots is the distinguished skills for many application areas. Planning gaits of efficient walking for quadruped robots is an important and challenging task. Especially, autonomous generation of locomotion is required to manage various robot models and environments. In this paper, we propose an adaptive locomotion control of 4 legged robot for irregular terrain using HyperNEAT. Generated locomotion is executed and analysed using ODE based Webots simulation for the 4 legged robot which is built by Bioloid.

Automatic Gait Generation for Quadruped Robot Using a GP Based Evolutionary Method in Joint Space

This paper introduces a new approach to develop a fast gait for quadruped robot using GP(genetic programming). Planning gaits for legged robots is a challenging task that requires optimizing parameters in a highly irregular and multidimensional space. Several recent approaches have focused on using GA(genetic algorithm) to generate gait automatically and shown significant improvement over previous results. Most of current GA based approaches used pre-selected parameters, but it is difficult to select the appropriate parameters for the optimization of gait. To overcome these problems, we proposed an efficient approach which optimizes joint angle trajectories using genetic programming. Our GP based method has obtained much better results than GA based approaches for experiments of Sony AIBO ERS-7 in Webots environment.

Shape, Volume Prediction Modeling and Identical Weights Cutting for Frozen Fishes

This paper suggests a modeling technique for shape and volume prediction of fishes to cut them with identical weights for group meals. The measurement and prediction of frozen fishes for group meals are very difficult because they have a bending deformation occurring at frozen stage and a hollow by eliminating the internals. Besides there exist twinkles problem of surface caused by freeze and variable weights by moisture conditions. Therefore a complex estimation algorithm is necessary to predict the shape and volume prediction of fishes exactly. Hollow prediction, pattern classification and modeling for tails using neural network, integration based volume prediction algorithm are suggested and combined to solve those problems. In order to validate the proposed method, the experiments of 3-dimensional measurement, volume prediction and fish cutting for spanish mackerel, saury, and mackerel are executed. The cutting experiments for real fish are executed.

Tree-structure-aware GP operators for automatic gait generation of quadruped robot

In this paper, we suggest tree-structure-aware GP operators that heed tree distributions in structure space and their possible structural difficulties. The main idea of the proposed GP operators is to place the generated offspring of crossover and/or mutation in a specified region of tree structure space insofar as possible, taking into account the observation that most solutions are found in that region. To enable that, the proposed operators are designed to utilize information about the region to which the parents belong and node/depth statistics of the subtree selected for modification. The proposed approach is applied to automatic gait generation of quadruped robot to demonstrate the effectiveness of it. The results show that the results using the proposed tree-structure-aware operators are superior to the results of standard GP for gait problem in both fitness and velocity.