AdaptiveCoPilot: Design and Testing of a NeuroAdaptive LLM Cockpit Guidance System in both Novice and Expert Pilots
DOI: 10.1109/vr59515.2025.00088
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper introduces AdaptiveCoPilot, a neuroadaptive guidance system designed to mitigate cognitive overload and underload in pilots by dynamically adjusting multimodal feedback based on real-time physiological data. The research addresses the challenge of managing high cognitive demands in modern cockpits, where complex interfaces and multitasking can lead to performance degradation and errors. Traditional guidance methods, such as static paper checklists and Crew Resource Management training, have not evolved significantly to address these dynamic cognitive states. The study aims to demonstrate that a system integrating functional Near-Infrared Spectroscopy (fNIRS) with a Large Language Model (LLM) can maintain optimal cognitive workload, thereby enhancing pilot performance and safety. The development of AdaptiveCoPilot involved a two-phase approach. First, a formative study with three expert pilots identified adaptive rules for modality switching and information load adjustments, leading to four key design requirements: preventing information overload through context-aware switching, dynamically adapting information load to manage complexity, providing timely multimodal feedback for error correction, and personalizing guidance based on pilot expertise. Second, the system was implemented in a high-fidelity virtual reality (VR) simulation of a UH-60 Black Hawk cockpit. The system uses fNIRS to monitor cortical hemodynamics in the prefrontal cortex, classifying cognitive states across working memory, perception, and attention into underload, optimal, or overload categories. These states, along with behavioral data, inform a quantized Microsoft PHI-3 LLM, which reasons via chain-of-thought to select appropriate guidance modalities (visual, auditory, textual) and information loads. The system was evaluated in a case study with eight licensed pilots, comparing the adaptive condition against baseline and random feedback conditions during preflight checklist tasks. The results indicate that pilots using AdaptiveCoPilot achieved higher rates of optimal cognitive load states in working memory and perception facets compared to baseline conditions. Additionally, the adaptive system led to reduced task completion times. The qualitative interviews with pilots from diverse backgrounds, including recreational, fighter, and Black Hawk pilots, highlighted the system's potential as a flexible safety assistant rather than a rigid command structure. Experts noted that multimodal feedback, particularly the combination of visual and auditory cues, was effective in correcting errors and maintaining engagement during routine tasks. The study also identified that isolating specific cognitive facets allows for more precise interventions, such as increasing information load during underload to prevent boredom or limiting it to essential commands during overload to reduce strain. The significance of this work lies in its proof-of-concept for generalized real-time neuroadaptive guidance in aviation. By demonstrating that fNIRS-driven LLM systems can effectively manage cognitive workload, the study provides a framework for future AI-adaptive assistants in high-stakes environments. The findings suggest that neuroadaptive systems can enhance pilot performance by maintaining optimal cognitive states, reducing errors, and improving situational awareness. The proposed strategies for designing such systems, including context-aware modality switching and dynamic information load adjustment, offer actionable insights for developing next-generation cockpit guidance tools that address the limitations of current static training methods.
Key finding
Pilots using the AdaptiveCoPilot system exhibited higher rates of optimal cognitive load states and reduced task completion times compared to baseline and random feedback conditions.
Methodology
simulator
Sample size: 8
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-27.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-27 |
| archive | success | canonical_url | — | — | 1 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | skipped | — | — | — | 3 | 2026-06-04 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.