TreeAgent: A Generalizable Multi-Agent Framework for Automated Bias Labeling in Forestry via Compiled Expert Rules and Vision-Language Models

Shiyi Chen, Nicholas Saban, Collin Hargreaves, Huiqi Wang · 2026

Read TreeAgent: A Generalizable Multi-Agent Framework for Automated Bias Labeling in Forestry via Compiled Expert Rules and Vision-Language Models by Shiyi Chen, Nicholas Saban, Collin Hargreaves, Huiqi Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-31976.txt.

Description

Chen and three colleagues automate a slow expert labeling task in forestry: classifying why tree height measurements disagree. Field crews, lidar and canopy height models each err differently, and experts sort individual trees into seven bias types using a rule-based diagnostic. Each tree takes three to five minutes of expert time, and experts disagree often enough to noise the reference data itself.

TreeAgent orchestrates compiled expert rules with vision-language models. A transpiler turns natural-language rules into validated decision graphs over a fixed inventory of primitives, so rule changes become configuration edits. Vision models handle only the perceptual nodes, with multi-agent voting to damp their randomness. The system beats supervised baselines on the testbed while cutting expert labeling effort.

Catalog record

CREATORShiyi Chen, Nicholas Saban, Collin Hargreaves, Huiqi Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.31976

Browse the full library · About the Tower of Babel