Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bidirectional Human-AI Learning in Real-Time Disoriented Balancing

Sheikh Mannan, Nikhil Krishnaswamy

TL;DR

The paper tackles spatial disorientation by enabling bidirectional learning between a human and an AI within a visually simulated inverted pendulum task (VIP). It employs a two-phase learning protocol—human training with AI-suggested cues and AI training refined by human corrections—assessed through phase portraits that capture mutual adaptation. Key contributions include a real-time, hardware-light demonstration with 26 AI assistants, a crash-prediction module for cueing, and public code availability to enable rapid exploration of human-AI trust and shared autonomy. The work offers a practical framework for studying dyadic human-AI interaction under disorientation, with broad relevance to piloting, spaceflight, and interactive AI systems.

Abstract

We present a real-time system that enables bidirectional human-AI learning and teaching in a balancing task that is a realistic analogue of disorientation during piloting and spaceflight. A human subject and autonomous AI model of choice guide each other in maintaining balance using a visual inverted pendulum (VIP) display. We show how AI assistance changes human performance and vice versa.

Bidirectional Human-AI Learning in Real-Time Disoriented Balancing

TL;DR

The paper tackles spatial disorientation by enabling bidirectional learning between a human and an AI within a visually simulated inverted pendulum task (VIP). It employs a two-phase learning protocol—human training with AI-suggested cues and AI training refined by human corrections—assessed through phase portraits that capture mutual adaptation. Key contributions include a real-time, hardware-light demonstration with 26 AI assistants, a crash-prediction module for cueing, and public code availability to enable rapid exploration of human-AI trust and shared autonomy. The work offers a practical framework for studying dyadic human-AI interaction under disorientation, with broad relevance to piloting, spaceflight, and interactive AI systems.

Abstract

We present a real-time system that enables bidirectional human-AI learning and teaching in a balancing task that is a realistic analogue of disorientation during piloting and spaceflight. A human subject and autonomous AI model of choice guide each other in maintaining balance using a visual inverted pendulum (VIP) display. We show how AI assistance changes human performance and vice versa.

Paper Structure

This paper contains 6 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Task Setting
  3. System Functionality
  4. Technical Specifics
  5. Conclusion
  6. Acknowledgments

Figures (2)

  • Figure 1: A human engaging in VIP balancing with cues (arrows) rendered by an AI assistant. A demonstration video can be viewed at https://youtu.be/coJdj0LIYa4.
  • Figure 2: Phase portraits of sample human VIP performance without [L] and with [R] AI assistance. With AI assistance, this human subject decreased their oscillation and maintained stability even while offset from the DOB.