Explore the power of gradients in solving inverse dynamics problems in this 54-minute talk by MIT's Tao Du. Delve into a broadened definition of inverse dynamics that infers computational design parameters for various dynamic systems. Learn about analytical gradients from physics simulators and their applications in robot design, digital twin construction, and real-world implementations. Discover how combining physics simulation, machine learning, and numerical optimization techniques can address challenges in computational fabrication, robotics, and machine learning. Gain insights into exploring shape and controller design spaces for rigid and soft robots, as well as transferring computational designs to hardware. Understand the simulation-to-reality gap of dynamic systems and the potential future opportunities in this field.
Overview
Syllabus
Introduction
Parameters
Forward Dynamics Problems
Inverse Dynamics Problems
Gradientbased Approaches
Problem Statement
Why Inverse Dynamics
Challenges
Results
Phase 2 Derived Ingredients
Development Goals
Examples
Trajectory Optimization
Sim Transfer Example
Challenge 1 Shadows
Challenge 1 Idea
Summary
Story
Questions
Taught by
Paul G. Allen School
