Rescue-Grid
An Open Platform for Human-Machine Teaming

From Workshop Prototype to Reusable Research Infrastructure

A tour of Rescue-Grid, a MiniGrid/Gymnasium-based search-and-rescue simulation built for studying human-AI teaming — covering its architecture, LSL/ixp research instrumentation, and how to extend it for your own study.

Authors

Affiliation

Part 1: Foundations ↗ Part 2: Case Study ↗

Overview

This workshop runs in two parts. Part 1: Foundations covers the principles of human-AI teaming — situational awareness, mental workload, and trust — and how each is measured, including with physiological sensing (eye tracking, EEG). Part 2: Case Study is a tour of Rescue-Grid, a MiniGrid/Gymnasium-based search-and-rescue simulation built for studying human-AI teaming under controlled, repeatable conditions, covering its architecture, LSL/ixp research instrumentation, and how to extend it for your own study — from a scripted RL environment to a full physiological experiment with eye-tracking and custom sensors.

Topics Covered

• Environment design — swappable camera views, real vs. decoy victims, and configurable difficulty as build_sar_env() parameters
• Trust and workload measurement — an optional LLM assistant whose suggestions are logged automatically as a behavioral trust signal
• LSL instrumentation — synchronizing game state with Tobii eye-tracking (or any registered sensor) on one shared clock
• ixp integration — how Rescue-Grid plugs into iHuman Lab’s experiment engine as a Task
• Extending the platform — swappable cameras, factories, and placers for building your own scenarios

Related Work

This platform produced a study on LLM-mediated communication in human-AI teaming, presented as LLM-Mediated Communication in Human–AI Teaming: A Physiological Monitoring Study for Search and Rescue (Oveisi & Manjunatha).

Venue

NeuroErgonomics 2026, Boston, MA — Wednesday, July 15, Session 6: Human–Robot & Human–AI Teaming