Autonomous robots are increasingly applied to tasks involving complex maneuvers and dynamic environments that are difficult to model a priori. Several types of learning approaches have been proposed to fill this modeling gap. To motivate the need for learning complex fluid-structure interactions, we first consider the SQUID (a ballistically launched and self-stabilizing drone) and PARSEC (an aerial manipulator capable of delivering self-anchoring sensor network modules) systems. We then show how to learn basic fluid-structure interactions using Koopman spectral techniques and incorporate the learned model into a real-time nonlinear model predictive control framework. The performance of this approach is demonstrated on small drones operating very close to the ground, where ground effect typically destabilizes flight. Operational risks abound in complex robotic tasks. This risk arises from both the uncertain environment and incompletely learned models. After discussing the basics of coherent risk metrics, we show how simple risk-aware terrain analysis improved the performance of our legged and wheeled robots in the DARPA Subterranean challenge. We then introduce an online method to learn the dynamics of an a priori unknown dynamical obstacle and robustly avoid the obstacle using a novel risk-based, distributionally robust, probability constraint derived from the evolving learned model. Next, we introduce the concept of risk surfaces to enable fast online learning of a priori unknown dynamical perturbations and show how this approach can adapt a drone to wind perturbations with only 45 seconds of online data collection.

0:00 Introduction
2:47 SQUID I: Key Design Elements
4:00 SQUID II: Vision-based autonomous stabilization
6:22 Planetary Exploration Applications
8:01 PARSEC: Robotic Load Anchoring System for Cliff Exploration Task Motivation and Description
8:32 PARSEC: Air Manipulator
9:23 Implementation Interface and Payload Design
10:06 Mission Architecture for Autonomous Deployment
11:39 But what about the real world?
12:30 Machine Learning & Nonlinear Vehicle Control
17:24 Using learned lifted bilinear models for nonlinear MPC
20:27 Learning from quadrotor dynamics to improve close-to-ground orbit tracking
21:00 Learning from quadrotor dynamics to improve close-to-ground trajectory following performance
24:00 Planning under uncertainty
24:56 Risk-aware planning: Opportunity limitations
28:15 The DARPA Underground Challenge
29:05 STEP: Stochastic Traversability Evaluation and Scheduling
34:07 Risk-conscious avoidance of unknown dynamic ostacles
42:54 Robust risk-based learning from disruptions
46:41 Learning and Introspective Control (LINC)

Please feel free to share this video with your friends and family if you found it useful.