TCOptRob Seminar: Carlos Mastalli and Majid Khadiv

TCOptRob Seminar: Carlos Mastalli and Majid Khadiv

Building Athletic Intelligence in Legged Robots: a Top-Down Approach by Carlos Mastalli of Heriot-Watt University (https://cmastalli.github.io/)

Optimal Control and Learning for Agile Locomotion by Majid Khadiv of the Max Planck Institute (https://is.mpg.de/~mkhadiv)

00:00 Intro
01:22 Carlos Mastalli: Building Athletic Intelligence in Legged Robots: a Top-Down Approach
27:17 Majid Khadiv: Optimal Control and Learning for Agile Locomotion
54:22 Q&A

Building Athletic Intelligence in Legged Robots: a Top-Down Approach

Building athletic intelligence has been a longstanding challenge in legged robotics. When generating agile motions, we push the robot's motor limit and require tracking angular momentum accurately. Ignoring these aspects is quite restrictive and does not hold for most robots (e.g., ANYmal B). From a top-down perspective, the main considerations when generating an agile manoeuvre are the robot's full dynamics and motor limits, which has implication for the robot's kinetic momenta, control and optimisation. In this talk, I will first share our efforts in developing advanced, efficient and open-sourced algorithms for feedback MPC with forward and inverse dynamics. This feedback MPC is a crucial element in our perceptive locomotion pipeline, which combines mixed-integer and quadratic programs for selecting optimal footstep regions and swing-leg motions that avoid obstacles in real-time. It enables the ANYmal robot to execute jumps, dynamics gaits and climb challenging obstacles. Then, I will also share our recent findings on differentiable optimal control for robotics and how these ideas are enabling simple ways to computationally design robots that can perform agile manurers. It enables us to compose codesign problems (a larger nonlinear program) while exploiting the temporal structure of optimal control. With this, we can interactively design the robot's limbs, motor sizes and distribution of components such as PCs and batteries.

Bio:
Carlos Mastalli is a robotics researcher working in the intersection of model predictive control and machine learning for motor control in legged robots. The aim of his research is to enable robots to move everywhere. In particular, he is focused on the problem of loco-manipulation tasks since it combines the main challenges in robot mobility. His research combines the formalism of model-base approach with the exploration of vast robot’s data.


Optimal Control and Learning for Agile Locomotion

Abstract:
Legged robots have become highly capable in the past few years, thanks to the rapid progress on both hardware and control software sides. However, there is still no consensus among researchers how to combine the strengths of control-theory-driven and data-driven approaches in different stages of the control pipeline. In this talk, I will present our recent efforts on developing algorithms that can benefit from best of both worlds for legged locomotion control.

Bio:
He is a research scientist in the Empirical Inference department led by Prof. Dr. Bernhard Schölkopf. His research focuses on both theoretical and empirical aspects of robotics, with a focus on locomotion and manipulation. He is mostly interested in generating complex motions for robots using machine learning and control theory and evaluating these behaviours on real robots. From May 2018 to June 2022, he was a postdoctoral researcher in the Movement Generation and Control Group led by Prof. Dr. Ludovic Righetti. During his postdoc, he also contributed to the European Project MEMMO as well as the developement of the Open Dynamic Robot Initiative (ODRI). Before joining Movement Generation and Control Group as a postdoc, he visited Autonomous Motion Department as a PhD visitor. During his one year visit, he worked on generating robust walking patterns for the humanoid robot Athena. From 2012 to 2015, he was the head of dynamics and control group in the Iranian national humanoid robot project, SURENA III. He designed some novel motion planning and control algorithms for humanoid robots and implemented on humanoid robot SURENA III.

----------------------------------

The IEEE RAS Technical Committee on Model Based Optimization for Robotics is focused on building and supporting a community of researchers and practioners focused on the development and application of model-based optimization techniques for the generation and control of dynamic behaviors in robotics and their practical implementation. You can find our more about the TC at: https://www.ieee-ras.org/model-based-....

Copyright (c) 2023 TCOptRob. All rights reserved.
License: CC BY-NC-SA 4.0