TY  - RPRT
TI  - Exact Action Values Are Not Enough: Rollout-Verified Reinforcement Fine-Tuning of a Reasoning Model for Multi-Zone VAV Control
AU  - Takumi Shioda
AU  - Kohei Terashima
AU  - Tatsuo Nagai
PY  - 2026
UR  - https://arxiv.org/abs/2607.27914
ID  - 2607.27914
ER  -