Non-Linear Control by Artificial Intelligence for complex and underactuated systems

The project is part of a PhD Thesis titled “Overcoming the Reality Gap: Imitation and Reinforcement Learning Algorithms for Bipedal Robotic Locomotion Problems” which aims to study the use of imitation and reinforcement learning algorithms to allow humanoid robots to learn how to walk in an efficient way. In particular, the project to be developed in Canada in the UQTR corresponds to the application of reinforcement learning algorithms for obtaining new gait patterns or for improving the efficiency of existing ones. This optimization process is performed realizing multiple tests of different robot controllers which tune their behavior based on a score called reward inspired in natural learning processes. These tests are realized through the iterative combination of simulations and experiments with the real robot that result in gait patterns that are fast and energetically efficient into the simulation environments as well as in the real robot.

Faculty Supervisor:

Alben Cárdenas

Student:

Partner:

Universidad Nacional de Colombia

Discipline:

Engineering

Sector:

Artificial Intelligence

University:

Université du Québec à Trois-Rivières

Program: