{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "Orca核心架构:统一世界潜空间+三目标预训练,backbone冻结只训readout validates 消融验证三组件独立可叠加贡献（预训练/后训练/RTC）", "strength": "medium", "meta": {"source_card_id": "new_fact_111", "target_card_id": "new_fact_032", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "contrastive_action_image_pre_training_for_visuomotor_control_2026_07", "relation_type": "same_track", "explanation": "Orca核心架构:统一世界潜空间+三目标预训练,backbone冻结只训readout validates 下游：冻结编码器→flow-matching策略，decoder用Qwen3.5-0.8B且从零训练", "strength": "medium", "meta": {"source_card_id": "new_fact_111", "target_card_id": "caip_fact_004", "target_paper": "Contrastive_Action_Image_Pre_training_for_Visuomotor_Control", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "modality_augmented_fine_tuning_of_foundation_robot_policies_for_cross_embodiment_2026_07", "relation_type": "same_track", "explanation": "Orca核心架构:统一世界潜空间+三目标预训练,backbone冻结只训readout validates 冻结骨干的轻量微调配置", "strength": "medium", "meta": {"source_card_id": "new_fact_111", "target_card_id": "new_fact_088", "target_paper": "Modality_Augmented_Fine_Tuning_of_Foundation_Robot_Policies_for_Cross_Embodiment", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "Orca核心架构:统一世界潜空间+三目标预训练,backbone冻结只训readout validates 联合co-training异构人机数据是结构性次优方案", "strength": "medium", "meta": {"source_card_id": "new_fact_111", "target_card_id": "new_boundary_031", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "动作生成对比:让Qwen3.5从0%成功率突破,且追平大规模机器人数据预训练的π0.5 validates Ψ0三阶段解耦训练范式", "strength": "strong", "meta": {"source_card_id": "new_fact_113", "target_card_id": "new_fact_025", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "contrastive_action_image_pre_training_for_visuomotor_control_2026_07", "relation_type": "same_track", "explanation": "动作生成对比:让Qwen3.5从0%成功率突破,且追平大规模机器人数据预训练的π0.5 validates 下游：冻结编码器→flow-matching策略，decoder用Qwen3.5-0.8B且从零训练", "strength": "medium", "meta": {"source_card_id": "new_fact_113", "target_card_id": "caip_fact_004", "target_paper": "Contrastive_Action_Image_Pre_training_for_Visuomotor_Control", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fast-wam_2026_07", "relation_type": "same_track", "explanation": "动作生成对比:让Qwen3.5从0%成功率突破,且追平大规模机器人数据预训练的π0.5 validates Fast-WAM无具身预训练即可达到SOTA接近水平", "strength": "medium", "meta": {"source_card_id": "new_fact_113", "target_card_id": "new_fact_014", "target_paper": "fast-wam", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "动作生成对比:让Qwen3.5从0%成功率突破,且追平大规模机器人数据预训练的π0.5 validates 联合co-training异构人机数据是结构性次优方案", "strength": "strong", "meta": {"source_card_id": "new_fact_113", "target_card_id": "new_boundary_031", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fe0_2026_07", "relation_type": "same_track", "explanation": "文本生成对比:Orca-4B相对Qwen3.5-4B提升非均匀,空间关系几乎持平 validates 开环聚合:0→100%跨具身数据使验证误差降约8%(−7.9%)", "strength": "medium", "meta": {"source_card_id": "new_fact_114", "target_card_id": "fe0_fact_004", "target_paper": "fe0", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fe0_2026_07", "relation_type": "same_track", "explanation": "文本生成对比:Orca-4B相对Qwen3.5-4B提升非均匀,空间关系几乎持平 validates 不均收益:L1/L2/L3受益最大,L4跟随,L5仅微弱改善", "strength": "medium", "meta": {"source_card_id": "new_fact_114", "target_card_id": "fe0_fact_006", "target_paper": "fe0", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates WMPO验证像素级世界模型的必要性", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "legacy_fact_017", "target_paper": "", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "qwen_robotworld_technical_report_unifying_embodied_world_modeling_through_langua_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates 跨域物理知识互补性：操作/驾驶/导航联合训练相互强化", "strength": "medium", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_claim_076", "target_paper": "Qwen_RobotWorld_Technical_Report_Unifying_Embodied_World_Modeling_through_Langua", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates General Flow的零样本迁移成绩", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "legacy_fact_008", "target_paper": "", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fast-wam_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates Fast-WAM无具身预训练即可达到SOTA接近水平", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_fact_014", "target_paper": "fast-wam", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "qwen_robotworld_technical_report_unifying_embodied_world_modeling_through_langua_2026_07", "relation_type": "extension", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 refines QWEN-ROBOTWORLD是状态转移世界模型，不是VLA策略执行器", "strength": "medium", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_boundary_080", "target_paper": "Qwen_RobotWorld_Technical_Report_Unifying_Embodied_World_Modeling_through_Langua", "votes": "2/3", "original_relation_type": "refines", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates Ψ0极端数据效率：800h+30h超越10倍数据量基线40%+", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_fact_026", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates 消融验证三组件独立可叠加贡献（预训练/后训练/RTC）", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_fact_032", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates Ψ0三阶段解耦训练范式", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_fact_025", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "qwen_robotworld_technical_report_unifying_embodied_world_modeling_through_langua_2026_07", "relation_type": "same_track", "explanation": "世界模型预训练可能是替代大规模机器人数据预训练的路径 validates 两阶段渐进训练：通用预训练→四阶段具身SFT", "strength": "strong", "meta": {"source_card_id": "new_claim_118", "target_card_id": "new_fact_075", "target_paper": "Qwen_RobotWorld_Technical_Report_Unifying_Embodied_World_Modeling_through_Langua", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "qwen_robotworld_technical_report_unifying_embodied_world_modeling_through_langua_2026_07", "relation_type": "same_track", "explanation": "作者展望:统一状态转移世界表征未来可扩展到AI for science等复杂系统 validates 跨域物理知识互补性：操作/驾驶/导航联合训练相互强化", "strength": "strong", "meta": {"source_card_id": "new_claim_119", "target_card_id": "new_claim_076", "target_paper": "Qwen_RobotWorld_Technical_Report_Unifying_Embodied_World_Modeling_through_Langua", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "qwen_robotworld_technical_report_unifying_embodied_world_modeling_through_langua_2026_07", "relation_type": "extension", "explanation": "作者展望:统一状态转移世界表征未来可扩展到AI for science等复杂系统 extends QWEN-ROBOTWORLD是状态转移世界模型，不是VLA策略执行器", "strength": "medium", "meta": {"source_card_id": "new_claim_119", "target_card_id": "new_boundary_080", "target_paper": "Qwen_RobotWorld_Technical_Report_Unifying_Embodied_World_Modeling_through_Langua", "votes": "2/3", "original_relation_type": "extends", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fe0_2026_07", "relation_type": "extension", "explanation": "作者展望:统一状态转移世界表征未来可扩展到AI for science等复杂系统 extends Fe0不是'堆数据就能解决物理泛化'——动作迁移触及硬天花板", "strength": "medium", "meta": {"source_card_id": "new_claim_119", "target_card_id": "fe0_boundary_001", "target_paper": "fe0", "votes": "2/3", "original_relation_type": "extends", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "unitachand_2026_07", "relation_type": "same_track", "explanation": "作者展望:统一状态转移世界表征未来可扩展到AI for science等复杂系统 validates MANO UV Map统一触觉表征是当前跨灵巧手形态迁移的可行路线,精细接触任务优于纯视觉/原始触觉对齐", "strength": "medium", "meta": {"source_card_id": "new_claim_119", "target_card_id": "claim_unitachand_20260627_001", "target_paper": "UniTacHand", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fast-wam_2026_07", "relation_type": "same_track", "explanation": "模型规模不够,语言/图像/动作三种读出能力之间存在此消彼长的权衡 validates Fast-WAM当前研究不涵盖大规模预训练和外层自回归循环", "strength": "strong", "meta": {"source_card_id": "new_boundary_120", "target_card_id": "new_boundary_016", "target_paper": "fast-wam", "votes": "3/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "fe0_2026_07", "relation_type": "same_track", "explanation": "模型规模不够,语言/图像/动作三种读出能力之间存在此消彼长的权衡 validates 难度梯队:感知(L1/L2)易,关系/规划(L3/L4)中,动作(L5)最难", "strength": "medium", "meta": {"source_card_id": "new_boundary_120", "target_card_id": "fe0_fact_005", "target_paper": "fe0", "votes": "2/3", "original_relation_type": "validates", "auto": true}}
{"source_paper_id": "orca_2026_07", "target_paper_id": "an_open_foundation_model_towards_2026_07", "relation_type": "same_track", "explanation": "消融实验:三个训练目标缺一不可,单独用观察目标最差(29.3),三者全用最好(48.0) validates 消融验证三组件独立可叠加贡献（预训练/后训练/RTC）", "strength": "medium", "meta": {"source_card_id": "new_fact_124", "target_card_id": "new_fact_032", "target_paper": "An_Open_Foundation_Model_Towards", "votes": "2/3", "original_relation_type": "validates", "auto": true}}