Smooth Gait Generation for Quadrupedal Robots Based on Genetic Algorithm Optimization

Gait generation is the process of finding a sequence of robot leg movements, which propel the robot in the desired direction when executed in a certain order. It is an optimization problem where multiple parameters need to be tuned in order to generate an optimal gait. In this paper, we propose a novel technique to improve the gait quality of a quadrupedal robot. In our proposed technique, we create an optimal fitness function for a Genetic Algorithm (GA) optimizer and use a trapezoidal velocity profile for joint movements. Our quadrupedal robot consists of 8 joints, 2 per leg. All joints are actuated by servo motors. The robot joints are controlled using a single layer Artificial Neural Network (ANN) whose inputs are the current robot joint angles and outputs are the target joint angles. The ANN is called every time the joints reach their target positions. A GA is used to optimize the ANN weights. The GA runs for a total of 100 generations over a population size of 10. The fitness function is a combination of the total distance traveled by the robot, and a scaling factor for the fitness value based on the overall joint movements. This discourages the GA from optimizing gaits that tend to an idle state. The controllers are selected based on how well they maximize the fitness function. The simulation of the robot is carried out in Open Dynamics Engine (ODE). The results show that the proposed technique considerably improves the overall fitness of the gait and the total distance traveled by the robot. Moreover, the proposed technique converges to an optimal gait in under 20 generations whereas the existing method takes over 40 generations. Furthermore, the robot joint movement is much smoother in the proposed method hence reducing the jerking in the robot motion.