Optimization-based Control of Legged Robots

7 giugno 2017
June 7, 2017
Contatti: 
DII - Dipartimento di Ingegneria Industriale
via Sommarive, 9 - 38123 Povo, Trento
Tel. 
+39 0461 282500 - 2503
Fax 
fax +39 0461 281977

Venue: Department of Industrial Engineering - Seminar room, via Sommarive 9, Povo - Trento - h: 10:30

  • Andrea Del Prete, LAAS/CNRS (Toulouse, France)

Abstract

The goal of my research is to empower legged robots with the ability to autonomously navigate unstructured environments. My approach to achieve this goal heavily relies on numerical optimization, and consists in formulating the problems of identification and control as optimization problems, and then using a numerical solver to find their solution. The advent of robust numerical solvers, coupled with fast CPUs, allows us today to solve complex optimization problems inside high-frequency control loops. This is one of the main reasons behind the great progress of legged robots during the last 10 years. In this talk I will summarize my contributions to this exciting field, which I applied to three legged robots: the small humanoid iCub, the hydraulic quadruped HyQ, and the humanoid HRP-2. I will start discussing the estimation of contact forces and joint torques on iCub, using tactile sensors and force/torque sensors. Then I will present my work on identification of friction and motor gain parameters for the implementation of joint torque control on iCub and HRP-2. This low-level controller provides the basis for the implementation of compliant whole-body control algorithms, such as Task-Space Inverse Dynamics (TSID), which I used to balance iCub and HRP-2, while making them compliant, hence safer in case of unexpected contact. I will discuss pros and cons of the implementation of Model Predictive Control on HRP-2 using a complete dynamic model of the system. Also, I will present a Quadratic-Programming based controller for high-slope terrain locomotion of quadruped robots. Finally, I will conclude showing more recent results on the application of robust optimization techniques to ensure the satisfaction of the robot constraints (e.g. force friction cones, joint limits, balance) despite the presence of uncertainties in the robot model, such as joint-torque tracking errors and inertial parameter errors.

Biography

Andrea Del Prete was born in Cesena (Italy) in 1984. He received his degree in Computer Engineering (with honors) from the 2nd faculty of the University of Bologna (Italy) in 2009. In March 2013 he got his PhD from the Cognitive Humanoids laboratory of the department of "Robotics Brain and Cognitive Sciences" in IIT, Genova. During his PhD Andrea had been working on estimation and control of contact forces on the small humanoid robot iCub, using distributed force and tactile sensors. Since 2014 he has been a PostDoc at LAAS-CNRS in Toulouse, working on compliant optimization-based control with HRP-2.

Download 
application/pdfOptimization-based Control of Legged Robots Seminar(PDF | 37 KB)