SharedKV-BT: Node-Local Typed Decisions for Behavior-Tree Agents

2026-10-05

AI Summary

This study addresses the high decision latency of autoregressive models and the limitations of shared-prefix methods that overlook state dependency and verification. To this end, we propose SharedKV-BT, a framework integrating behavior tree architectures with a Shared-KV parallel scoring mechanism to enable efficient, controlled decision-making via node-local fields. Furthermore, phase gating is introduced to prevent out-of-order action execution, while external postconditions are leveraged to avoid premature task termination, thereby balancing execution efficiency with logical correctness. Experimental results demonstrate that the proposed system accelerates decision-making by 2.36× to 4.15×, achieving a joint accuracy of 94% on manipulation tasks and a closed-loop success rate of 60%.

Abstract

Agent tasks require sequences of interdependent decisions. Autoregressive models support more flexible decision interfaces than conventional classifiers but incur the latency of token-by-token generation. Recent shared-prefix methods reduce this cost by reusing encoded context and scoring multiple decisions in parallel, but do not model decision dependencies or verify execution. We propose SharedKV-BT, where each active node of a behavior tree (BT) exposes stage-local fields and candidates, and Shared-KV scores the candidates in parallel and passes the selected decision to a separate execution system. We tested SharedKV-BT on robot manipulation, mobile navigation, and computer-use tasks. Across three tasks, SharedKV-BT made typed decisions 2.36-4.15 times faster than prompt-matched autoregressive decoding. On the manipulation task, node-local Shared-KV improved joint decision accuracy from 75% to 94% and closed-loop success from 0% to 60%. Fixed-score policy replay showed that stage gating prevented out-of-order actions and external postconditions prevented premature completion.

Problem

• Agent decision-making
• Autoregressive latency
• Shared-prefix methods
• Decision dependencies
• Execution verification

Innovation

• SharedKV-BT
• Behavior Tree
• Shared-KV Cache
• Typed Decisions
• Parallel Scoring