MACHINE LEARNING IN THE GENERATION OF GAITS FOR LEGGED ROBOTS

Locomotion is an important component of mobile robots. Wheeled locomotion is fast and easy to implement but unfortunately a large portion of our environment is inaccessible to wheeled robots. While legged locomotion may also be fast, as evidenced by the cheetah's top speed of around 70 mph, even the simplest implementations of legged gaits require extensive tuning and testing to maintain stability and propel the robot forward. An effective approach to optimizing gaits is to use autonomous learning which allows a large space of possible gaits to be searched with minimal human interaction. This survey provides an overview of different robotic subsystems and shows how robot locomotion may be classified based on the number of legs or wheels of the robot. Different methods of learning are explored and classified based on how they define the gait parameters, what learning method is used, and how the gait is evaluated.