Explore cutting-edge developments in autonomous robotics for dynamic tasks in this Stanford seminar featuring Joel Burdick from Caltech. Delve into innovative projects like SQUID, a self-stabilizing drone, and PARSEC, an aerial manipulator for sensor network deployment. Learn about advanced techniques in fluid-structure interaction modeling using Koopman spectral methods and their integration into real-time nonlinear model predictive control. Discover how risk-aware approaches enhance robot performance in uncertain environments, from terrain analysis in the DARPA Subterranean Challenge to obstacle avoidance using distributionally robust chance constraints. Gain insights into fast online learning of dynamical disturbances through risk surfaces, enabling drones to adapt to wind conditions rapidly. This comprehensive talk covers a wide range of topics, including machine learning in vehicle control, planning under uncertainty, and introspective control, providing a thorough overview of the latest advancements in robotic systems for complex, dynamic environments.
Overview
Syllabus
Introduction
SQUID I: Key Design Elements
SQUID II: Vision-based Autonomous Stabilization
Planetary Exploration Applications
PARSEC: Payload Anchoring Robotic System for the Exploration of Cliffs Task Motivation and Description
PARSEC: Aerial Manipulator
Deployment Interface and Payload Design
Mission Architecture for Autonomous Deployment
But what about the real world?
Machine Learning & Nonlinear Vehicle Control
Using learned lifted bilinear models for nonlinear MPC
Learning quadrotor dynamics to improve close-to-ground trajectory tracking
Learning quadrotor dynamics to improve close-to- ground trajectory tracking performance
Planning under uncertainty
Risk-Aware Planning: Chance Constraints
The DARPA Subterranean Challenge
STEP: Stochastic Traversability Evaluation and Planning
Risk-Aware Avoidance of Unknown Dynamic Ostacles
Robust Risk-Based Learning of Disturbances
Learning and Introspective Control LINC
Taught by
Stanford Online
