Synthesizing Adversarial Visual Scenarios for Model-Based Robotic Control

TL;DR

This paper introduces a method to generate adversarial visual scenarios specifically for multi-step, model-based robotic control, improving robustness of perception models against out-of-distribution inputs.

Contribution

It presents a novel approach to synthesize adversarial scenarios tailored for multi-step control, leveraging differentiable MPC to enhance robustness of vision models in robotics.

Findings

Re-training on adversarial datasets improves control performance by up to 36.2%.

Method demonstrated on navigation, manipulation, and autonomous air vehicle control.

Synthesizes adversarial scenarios specific to multi-step control tasks.

Abstract

Today's robots often interface with data-driven perception and planning models with classical model-predictive controllers (MPC). Often, such learned perception/planning models produce erroneous waypoint predictions on out-of-distribution (OoD) or even adversarial visual inputs, which increase control costs. However, today's methods to train robust perception models are largely task-agnostic - they augment a dataset using random image transformations or adversarial examples targeted at the vision model in isolation. As such, they often introduce pixel perturbations that are ultimately benign for control. In contrast to prior work that synthesizes adversarial examples for single-step vision tasks, our key contribution is to synthesize adversarial scenarios tailored to multi-step, model-based control. To do so, we use differentiable MPC methods to calculate the sensitivity of a model-based controller to errors in state estimation. We show that re-training vision models on these adversarial datasets improves control performance on OoD test scenarios by up to 36.2% compared to standard task-agnostic data augmentation. We demonstrate our method on examples of robotic navigation, manipulation in RoboSuite, and control of an autonomous air vehicle.