Generate-then-Rank：提升具身智能体符号目标理解的两阶段方法
Generate-then-Rank: Learning Symbolic Goal Interpretation for Embodied Agents
把自然语言指令翻译成规划器能执行的符号目标，用微调加自排序两步提升准确率，做具身智能体的可以看看论文里的 CORE 方法。
一篇 arXiv 论文针对具身智能体的目标理解任务，即把自然语言指令转化为规划器可执行的符号化目标，提出两阶段的 Generate-then-Rank 方法。第一阶段用监督微调让 LLM 适配 Embodied Agent Interface（EAI）的输出格式，第二阶段提出 CORE 算法，从模型自生成的候选输出中学习偏好排序。在 EAI Benchmark 家庭场景上的实验显示，该方法超越了简单提示、思维链和知识图谱等基线。论文评测了 DeepSeek-V3 chat、DeepSeek-Reasoner、Qwen2.5-7B 和 Qwen2.5-72B 多个开源模型。
Generate-then-Rank: Learning Symbolic Goal Interpretation for Embodied Agents
Goal interpretation is to translate a natural language instruction into a formal symbolic goal that a planner can execute. It is critical for embodied agents because any error here propagates to every downstream planning and action step. Doing it well requires both generating suitable candidates and reliably selecting among them. We present a two-stage generate-then-rank approach to Goal Interpretation in the Embodied Agent Interface (EAI). First, supervised fine-tuning adapts a Large Language Model (LLM) to EAI's output convention. Second, Candidate-Output Ranking from Errors (CORE) learns preferences over complete outputs from the model's own generated alternatives. We conducted experiments on the household scenes of the Embodied Agent Interface Benchmark. Multiple open source LLMs including DeepSeek-V3 chat, DeepSeek-Reasoner, and Qwen2.5-7B and Qwen2.5-72B were evaluated. The proposed method outperformed the baseline models including LLMs with simple prompting, chain-of-thought, and knowledge graph based approaches.
- DAIR.AI11-04 12:35原文
