A Review of Gait Optimization Based on Evolutionary Computation

Abstract

Gait generation is very important as it directly affects the quality of locomotion of legged robots. As this is an optimization problem with constraints, it readily lends itself to Evolutionary Computation methods and solutions. This paper reviews the techniques used in evolution-based gait optimization, including why Evolutionary Computation techniques should be used, how fitness functions should be composed, and the selection of genetic operators and control parameters. This paper also addresses further possible improvements in the efficiency and quality of evolutionary gait optimization, some problems that have not yet been resolved and the perspectives for related future research.

Towards automatic discovery of agile gaits for quadrupedal robots

“…A variety of cost terms can be defined (see [22] for a comprehensive survey). For instance, to promote energetically efficient gaits, the collision losses, positive mechanical work, or the square of joint torques [23] can be minimized.…”

Comparison of a greedy selection operator to tournament selection and a hill climber

“…Formula (1) was used in generating the lists of greedy factors; Formula (2) calculates a greedy fitness from an actual fitness value and its corresponding greedy factor. f g (i) = i a (1) F g (i) = F(i) * f g (i) (2) In the formula fg(i) is the greedy factor corresponding to the position i in the fitness list, Fg(i) is greedy fitness, F(i) is an actual fitness score, i is an integer 1 to 64 representing fitness position in the population of size 64, and a is a greedy constant -a real number. For the test reported in this paper, five greedy constants (a) were used, 0 (equivalent to the standard roulette wheel), 0.2, 0.5, 1 and 1.5 were used in generating the list of greedy factors depending on the value of i.…”

“…Daoxing Gong et al [1], in a recent article, reviewed the suitability of evolutionary computation techniques in gait optimization for mobile legged robots. In a previous work [2], Parker et al used Cyclic Genetic Algorithms (CGAs) to generate gaits for legged robots using minimal a priori knowledge.…”

Quadruped gait learning using cyclic genetic algorithms

“…In a recent review article, Daoxiong Gong, Jie Yan, and Guoyu Zuo presented, in much detail, the suitability of evolutionary computation techniques in gait optimization for mobile legged robots [1]. Graham Spencer used genetic programs in his work to learn gaits for a virtual robot using only minimal knowledge of the mechanisms of walking [2].…”