Human-AI Collaboration in Autonomous Aerial Vehicles
Abstract
Looking 10-20 years into the future, in the rapidly advancing world of autonomous, special ops, cargo, and medevac aircraft, all of the basic aviation functions will likely be handled completely and competently by AI agents embedded within air vehicles. Under nom inal conditions (and many basic off-nominal situations) the AI-controlled vehicle will operate autonomously and independently withou t input from on-board personnel. However, no mission is ever completely nominal, and many open questions remain about how on-board p ersonnel and AI controlling the vehicle should collaborate effectively. Fluency is the quality of interaction between a human and a robot and has been used to evaluate many aspects of human-robot teaming. In this project, we ask: What is the impact of Human-AI te aming, i.e. fluency on mission effectiveness, and how canit be fostered and maintained?The goal of this research is to enable appr opriate human-AI collaboration needed to deal with off nominal events by (1) characterizing the challenges to fluency created by hum an biases and cognitive limitations as they impact Human-AI interaction, (2) quantifying the impact of fluency on mission effectiven ess, and (3) exploring and validating mitigation strategies. Specifically we seek to understand elements of fluency that are needed for an AI agent to seek and receive assistance from on-board personnel who have no direct training in piloting or AI programming. W hen devising mitigation strategies, our focus will be on mitigations that can be employed dynamically in response to the operators behaviors or cognitive state, or in response to drops in fluency. By deploying changes to the AI system and how it engages with the human team member, the AI system can mimic positive attributes in human teaming whereby members change theirbehavior toward one an other based on context and assessment of what is needed to achieve the mission goals.The proposed work is divided into three resear ch phases with multiple sub-tasks in each: (1) Establishment of Realistic Scenarios and Models, (2) Characterization of Human-AI Col mental scenarios that exercise and span the operational scenarios of interest, induce human biases and cognitive limitations, and pr oduce realistic autonomous responses to these situations including realistic flight dynamics and trade-offs between performance, pas senger comfort, and mission success. In the second phase we will seek to understand Human-AI collaboration behaviors in each of the scenarios developed previously. We will rely primarily onempirical evaluations and seek to understand when the identified human bi ases are likely to arise based on scenario parameters and to develop metrics that will allow us to characterize the Human-AI collabo ration effectiveness in real-me. In the final research phase we will create and evaluate potential mitigation strategies designed to improve mission effectiveness and teaming fluency by helping humans to appropriately calibrate their trust in the AI pilot and help ing the AI pilot to be more robust to the variance inherent in human collaboration. d our knowledge of Human-AI interaction to a broader class of tasks which requiredistinct interaction methods, (2) establish a mean s for providing real me assessments of human behaviors and cognitive states with respect to their interaction with the AI pilot, and (3) extend fluency measures for the real-me assessment of Human-AI teaming in order to establish formal guarantees of operational s afety and performance. As a result, we will have a greater understanding of how different task characteristics influence Human-AI te aming and fluency.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 20, 2021
- Source ID
- N000142112759
Entities
People
- Karen Feigh
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy