22 Março, 2017
Em parceria com a IEEE Seção Centro-Norte Brasil, o Programa de Pós-Graduação em Engenharia de Sistemas Eletrônicos e de Automação (PGEA) da UnB gostaria de convidar todos a participarem da seguinte palestra “Optimization and Learning for Planning and Control of Legged Robots“.
Palestrante: Dr. Diego Pardo
Local: Sala de Aula Sérgio Barroso – CDT (Prédio da Engenharia Elétrica, Faculdade de Tecnologia, UnB)
The field of planning and control for legged robots has progressed extraordinarily over the last few years. Methods to make robots walk, run and jump exist and they are well understood. However, such methods are usually specific for a given task and robot morphology. Generalized methods are typically based on simplified models, and advanced robots are not using all their dynamic capabilities to move efficiently, but their motions are driven by the premise of ‘not to fall.’ The current challenge is to improve its performance (e.g., speed, versatility, robustness) and to find generic frameworks able to generate diverse types of dynamic motions from simple specifications.
In this talk, I will present the research conducted in the Agile and Dexterous Robotics Lab at ETH Zürich towards the motion planning and control of legged robots. We have addressed this problem using numerical optimization and learning approaches, and we have developed solutions from different perspectives: ranging from very fast on-line optimization of simplified models to whole-body trajectory optimization through contacts. We have applied these methods on a hydraulically actuated and torque controlled quadruped robot, demonstrating different gaits and aperiodic dynamic motions in the real system.
Sobre o palestrante
Agile & Dexterous Robotics Lab – ETH Zürich