UW CSE Robotics - Emo Todorov, "Goal-Directed Dynamics"

Explore goal-directed dynamics in robotics through this lecture by Emo Todorov. Delve into a general control framework that integrates a low-level optimizer into robot dynamics, creating a goal-directed dynamical system controlled at a higher level. Discover how this approach uses cost functions to encode desired accelerations, end-effector poses, and center of pressure. Learn about the non-convex and non-smooth optimization problem solved at each time step, and how it leverages the unique properties of a soft-constraint physics model. Gain insights into the potential applications of this computational infrastructure, including tele-operation, feature-based control, deep learning of control policies, and trajectory optimization. Understand the syllabus covering topics such as robotics automation, model-based optimization, online trajectory optimization, learning progress on hardware platforms, notation, dynamic programming, and future work in the field.

Intro
Robotics in need of automation
Model-based optimization
Online trajectory optimization (MPC)
Learning progress (hardware platform)
Learned behaviors (hardware platform)
Notation
Definition of goal-directed dynamics
Dynamic Programming
Future work