BASL: Grounding LLM-Based Historical Battle Simulation Using Execution Difficulty and Tactical Necessity

BattleAgent (BA) uses large language models (LLMs) to select tactical Actions flexibly, but LLM decisions can conflict with the simulation state, execution rules, or tactical situation. We divide this LLM–simulation grounding gap into two components, the Action-execution grounding gap and the tactical Action-selection grounding gap, and introduce simulation grounding fidelity (SGF) to assess them. We propose BASL (BattleAgent with a State Ledger), in which the simulator enforces Action feasibility, calculates execution Difficulty D(a,St) and tactical Necessity N(a,St) from the current state, and provides them to the LLM as advisory information while leaving the final Action choice to the LLM. Experiments used four battlefields: Crécy, Poitiers, Agincourt, and Falkirk. Presenting both D and N raised the tactical-grounding index SGFT from 65.105 to 67.682. Across four LLMs, presenting Difficulty reduced mean selected Difficulty, whereas presenting Necessity increased mean selected Necessity. In 50 Agincourt simulations, the simulator’s checks found no violations of movement distance, weapon range, or Action feasibility. Thus, presenting D and N together with instructions to prefer lower-risk and sufficiently necessary Actions shifts Action selection in their direction, while BASL executes Actions according to the simulator’s rules. We do not claim that the resulting decisions are tactically optimal or better than those of human commanders.

Authors

Institutions

Publication Details

Journal
Electronics
Published
2026-11-17
DOI
https://doi.org/10.3390/electronics15225364
Primary Topic
Artificial Intelligence in Games
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

BASL: Grounding LLM-Based Historical Battle Simulation Using Execution Difficulty and Tactical Necessity

Hiroshi Satō, Masao Kubo, Akihiro Yamaguchi, Ryugo Okuyama
5 links
Electronics
Artificial Intelligence in Games
article

BASL: Grounding LLM-Based Historical Battle Simulation Using Execution Difficulty and Tactical Necessity

Hiroshi Satō, Masao Kubo, Akihiro Yamaguchi, Ryugo Okuyama
article en

Abstract

BattleAgent (BA) uses large language models (LLMs) to select tactical Actions flexibly, but LLM decisions can conflict with the simulation state, execution rules, or tactical situation. We divide this LLM–simulation grounding gap into two components, the Action-execution grounding gap and the tactical Action-selection grounding gap, and introduce simulation grounding fidelity (SGF) to assess them. We propose BASL (BattleAgent with a State Ledger), in which the simulator enforces Action feasibility, calculates execution Difficulty D(a,St) and tactical Necessity N(a,St) from the current state, and provides them to the LLM as advisory information while leaving the final Action choice to the LLM. Experiments used four battlefields: Crécy, Poitiers, Agincourt, and Falkirk. Presenting both D and N raised the tactical-grounding index SGFT from 65.105 to 67.682. Across four LLMs, presenting Difficulty reduced mean selected Difficulty, whereas presenting Necessity increased mean selected Necessity. In 50 Agincourt simulations, the simulator’s checks found no violations of movement distance, weapon range, or Action feasibility. Thus, presenting D and N together with instructions to prefer lower-risk and sufficiently necessary Actions shifts Action selection in their direction, while BASL executes Actions according to the simulator’s rules. We do not claim that the resulting decisions are tactically optimal or better than those of human commanders.

ElectronicsVol. 15(22)
Fukuoka Institute of Technology (JP), National Defense Academy of Japan (JP)
Asian Office of Aerospace Research and Development
Artificial Intelligence in Games
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon