评测档案

SafeAgentBench

具身规划器能否识别危险指令，同时完成安全任务？

本轮尚未登记机构署名 / 来源捕获 · fc3f276d7f / Behavior · 观察行为

来源与出处 ↓ 阅读研究记录 →

跨报告监测

已发表证据

SafeAgentBench · Rejection ratio

跨来源报告的已发表测量。任务范围通过注释区分，配置差异保留在每条观测中。

180 条观测 · 0 已定位日期

● 条件匹配◇ 配置差异□ 部分条件未披露

Detailed hazardous tasks
Abstract hazardous tasks
Abstract tasks / L1
Abstract tasks / L2
Abstract tasks / L3
Abstract tasks / L4

180 条观测尚无已核实的模型发布日期，可在数据与系统结果中查看。

选择图中数据点或表格行，查看条件与出处。

180 条观测

测量值与来源条件
系统／模型	结果	条件状态	任务／条件	评测方／报告	日期
GPT-4 / LLM-Planner	0.63 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0.63 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0.56 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0.55 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.48 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / LLM-Planner	0.4 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0.36 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.35 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.32 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / LLM-Planner	0.31 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0.31 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.28 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.27 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0.26 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0.24 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.24 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.2 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.2 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.2 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.19 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.19 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0.17 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0.16 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.15 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0.15 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.13 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0.13 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0.13 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / LLM-Planner	0.12 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0.11 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ReAct	0.1 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0.1 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / LLM-Planner	0.1 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0.1 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0.09 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0.09 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0.09 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0.08 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0.08 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.07 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / ProgPrompt	0.07 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0.07 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0.07 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0.06 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.05 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0.05 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MLDT	0.05 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0.03 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0.03 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0.03 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0.03 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0.03 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0.02 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0.02 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0.02 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0.01 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0.01 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0.01 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / LLM-Planner	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / CoELA	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / CoELA	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / MLDT	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / CoELA	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / CoELA	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ProgPrompt	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ProgPrompt	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Detailed hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract hazardous taskstemperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / CoELA	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)CoELA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / MAP	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)MAP adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
GPT-4 / KARMA	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)KARMA adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ProgPrompt	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ProgPrompt adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Llama3-8B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / MLDT	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
Qwen2-7B / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Lota-Bench	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Lota-Bench adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / LLM-Planner	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)LLM-Planner adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / Co-LLM-Agent	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)Co-LLM-Agent adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / MLDT	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)MLDT adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / ReAct	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)ReAct adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L1temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L2temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L3temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17
DeepSeek-V2.5 / PCA-EVAL	0 ratioRejection ratio	条件不完整	Abstract tasks / L4temperature=0; max_tokens=4096 (Appendix C.1)PCA-EVAL adapted to SafeAgentEnv	未披露SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents	发布 —报告 2024-12-17

任务与测量

Benchmark 协议档案

任务／对象: 家庭场景模拟区分具体、抽象和长程任务，以执行结果和语义判断进行互补测量。
测量目标: 具身规划器能否识别危险指令，同时完成安全任务？
评分／结果口径: Rejection ratio · Rejection ratio
解释边界: 底层控制器将高层动作映射到模拟器接口；结果不能直接验证实体机器人或学习型运动控制器。本笔记不转录危险任务操作。

一项已报告评测

来源评测 GPT-4 / Lota-Bench · Detailed hazardous tasks.

运行配置: Lota-Bench adapted to SafeAgentEnv · temperature=0; max_tokens=4096 (Appendix C.1).

报告指标: Rejection ratio · 0 ratio.

Rejection ratio 见原文样本说明

修订版本报告不同规划方法虽有任务完成差异，危险识别整体仍弱；仅替换基础模型未可靠解决问题。

测量档案 · NUWA 解读

档案区分来源方法阅读与既有目录信息；已经公开不代表完成独立证据复核或双语编辑复核。

测量构念: 具身任务规划的安全表现，区分拒绝、风险执行与安全完成。
任务与样本: 包含 750 项模拟具身任务，其中 450 项有危险、300 项为对应安全任务；详细、抽象与长时程任务分别测量不同规划要求。
判分与失效情形: 详细任务报告拒绝率、基于目标／LLM 的风险率及执行率；抽象任务报告拒绝与风险率；长时程区分安全完成、不安全完成与未完成。比例保留原文 0–1 尺度。
系统条件: 各基线适配到 SafeAgentEnv。附录 C.1 设置 temperature=0、max_tokens=4096，C.2 列出 17 项高层动作；GPT-4 主表与三种开源模型附录结果分别保留。
统计含义: 现有材料尚未确定
效度与解释: 模拟器任务结果不能直接解释为现实物理事件频率；任务类型、判分方式和智能体设计都会改变比例含义。
适用研究问题: 现有材料尚未确定

SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents · #S3; #S4; #S5fc3f276d7f8003ffcc110c585a40c0ef6f2b88207d4da50f060d4d4ee32b4031

跨报告证据

相同 Benchmark 名称可能对应不同任务、计分规则和系统配置。各来源分别保留，比较组说明具体对齐了哪些条件。

SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents

实际评测方: 未披露

540 条结果记录 · 32 个报告系统名称

协议与比较依据 (0)

原始来源图表与材料

阅读原始协议 ↗

底层控制器将高层动作映射到模拟器接口；结果不能直接验证实体机器人或学习型运动控制器。本笔记不转录危险任务操作。

参考文献

数值结果由所引研究的作者报告。

[1] SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents. 2024-12-17 ; #S4.T2; row GPT-4 / Lota-Bench; condition Detailed hazardous tasks; column Rejection ratio; #S4.T2; row GPT-4 / Lota-Bench; condition Detailed hazardous tasks; column Risk ratio (goal evaluator)

搜索 NUWA Index