{"paper_id": "what_matters_for_latent_actions_in_robot_learning_2026_08", "slug": "what_matters_for_latent_actions_in_robot_learning", "title": "潜在动作:光流是负资产", "post_url": "https://haiguangboy.com/posts/what_matters_for_latent_actions_in_robot_learning", "asset_base_url": "https://static.haiguangboy.com/papers/what_matters_for_latent_actions_in_robot_learning", "tags": ["embodied_ai", "action_generation", "robotics", "state_prediction", "具身智能", "潜在动作", "表示学习"], "importance_score": 3.0, "ruling": "潜在动作:光流是负资产", "published_at": "2026-09-01T08:12:56+08:00", "updated_at": "2026-09-03T12:16:52+08:00"}
{"paper_id": "skild_blog_introducing_s1_in_context_learning_20260817_2026_08", "slug": "skild_blog_introducing_s1_in_context_learning_20260817", "title": "S1:机器人的GPT-3涌现时刻", "post_url": "https://haiguangboy.com/posts/skild_blog_introducing_s1_in_context_learning_20260817", "asset_base_url": "https://static.haiguangboy.com/papers/skild_blog_introducing_s1_in_context_learning_20260817", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "具身智能", "机器人基础模型", "上下文学习"], "importance_score": 3.0, "ruling": "S1:机器人的GPT-3涌现时刻", "published_at": "2026-08-26T22:25:45+08:00", "updated_at": "2026-09-03T12:14:54+08:00"}
{"paper_id": "pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation_2026_08", "slug": "pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation", "title": "PointWorld:点流统一状态动作", "post_url": "https://haiguangboy.com/posts/pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation", "asset_base_url": "https://static.haiguangboy.com/papers/pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "state_prediction", "具身智能", "世界模型", "3d点云"], "importance_score": 3.0, "ruling": "PointWorld:点流统一状态动作", "published_at": "2026-08-26T07:55:46+08:00", "updated_at": "2026-09-03T12:13:14+08:00"}
{"paper_id": "the_missing_touch_spatially_distributed_tactile_feedback_brings_teleoperation_cl_2026_08", "slug": "the_missing_touch_spatially_distributed_tactile_feedback_brings_teleoperation_cl", "title": "指尖麻醉类比:遥操作缺的那味触觉", "post_url": "https://haiguangboy.com/posts/the_missing_touch_spatially_distributed_tactile_feedback_brings_teleoperation_cl", "asset_base_url": "https://static.haiguangboy.com/papers/the_missing_touch_spatially_distributed_tactile_feedback_brings_teleoperation_cl", "tags": ["embodied_ai", "action_generation", "robotics", "state_prediction", "具身智能", "遥操作", "触觉反馈"], "importance_score": 3.0, "ruling": "指尖麻醉类比:遥操作缺的那味触觉", "published_at": "2026-08-25T18:41:14+08:00", "updated_at": "2026-09-03T12:16:22+08:00"}
{"paper_id": "flex_pi_a_multi_stream_world_action_model_with_compute_flexibility_2026_08", "slug": "flex_pi_a_multi_stream_world_action_model_with_compute_flexibility", "title": "把RGB、3D几何、物体语义揉进同一个表征——这才是它比π0.5、Fast-WAM更准的根本原因", "post_url": "https://haiguangboy.com/posts/flex_pi_a_multi_stream_world_action_model_with_compute_flexibility", "asset_base_url": "https://static.haiguangboy.com/papers/flex_pi_a_multi_stream_world_action_model_with_compute_flexibility", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "具身智能", "世界模型", "机器人操作"], "importance_score": 3.0, "ruling": "FLEX-π:更全面的联合表征", "published_at": "2026-08-25T09:01:40+08:00", "updated_at": "2026-09-03T12:09:22+08:00"}
{"paper_id": "prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim_2026_08", "slug": "prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim", "title": "PRISM:规模能否替代显式建模?", "post_url": "https://haiguangboy.com/posts/prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim", "asset_base_url": "https://static.haiguangboy.com/papers/prism_precision_and_contact_rich_real_world_industrial_skill_dataset_with_multim", "tags": ["embodied_ai", "action_generation", "robotics", "state_prediction", "具身智能", "机器人数据集", "工业操作"], "importance_score": 3.0, "ruling": "PRISM:规模能否替代显式建模?", "published_at": "2026-08-24T14:24:27+08:00", "updated_at": "2026-09-03T12:13:26+08:00"}
{"paper_id": "hydra-0-action-flow_2026_08", "slug": "hydra-0-action-flow", "title": "英伟达的动作流：机器人动作新标准", "post_url": "https://haiguangboy.com/posts/hydra-0-action-flow", "asset_base_url": "https://static.haiguangboy.com/papers/hydra-0-action-flow", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "state_prediction", "具身智能", "世界模型", "论文分享"], "importance_score": 3.0, "ruling": "英伟达把机器人动作定义成\"动作流\"——机械臂、人手、夹爪全部通用，误差直接降了90.40%", "published_at": "2026-08-22T11:34:36+08:00", "updated_at": "2026-09-03T12:10:06+08:00"}
{"paper_id": "gen-1-5-one-shot_2026_08", "slug": "gen-1-5-one-shot", "title": "机器人GPT-3时刻：故事很足，证据不够", "post_url": "https://haiguangboy.com/posts/gen-1-5-one-shot", "asset_base_url": "https://static.haiguangboy.com/papers/gen-1-5-one-shot", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "具身智能", "世界模型", "论文分享"], "importance_score": 3.0, "ruling": "Generalist AI说GEN-1.5不用任何专门设计就学会单样本操作——官方自己写的原文是：10个任务，方差±10%，技能\"比微调更脆弱\"", "published_at": "2026-08-21T17:40:24+08:00", "updated_at": "2026-09-03T12:09:54+08:00"}
{"paper_id": "omega-eva_2026_08", "slug": "omega-eva", "title": "世界模型：推理时到底要不要想象", "post_url": "https://haiguangboy.com/posts/omega-eva", "asset_base_url": "https://static.haiguangboy.com/papers/omega-eva", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "世界模型", "论文分享"], "importance_score": 3.0, "ruling": "SLIM、JEPA-WAM都说别在推理时干活——这篇消融显示：去掉那一步，精化器反而比不精化还差", "published_at": "2026-08-19T15:29:31+08:00", "updated_at": "2026-09-03T12:12:18+08:00"}
{"paper_id": "slim-05b_2026_08", "slug": "slim-05b", "title": "预测误差更低，表征却已经塌了", "post_url": "https://haiguangboy.com/posts/slim-05b", "asset_base_url": "https://static.haiguangboy.com/papers/slim-05b", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "世界模型", "论文分享"], "importance_score": 3.0, "ruling": "世界模型会预测未来，不等于会控制——给动作能预测后果，给后果能反推动作，只用 0.47B", "published_at": "2026-08-14T08:20:39+08:00", "updated_at": "2026-09-03T12:15:06+08:00"}
{"paper_id": "jepa-wam_2026_08", "slug": "jepa-wam", "title": "JEPA-WAM：创新方法让世界模型更懂状态变化", "post_url": "https://haiguangboy.com/posts/jepa-wam", "asset_base_url": "https://static.haiguangboy.com/papers/jepa-wam", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "世界模型"], "importance_score": 3.0, "ruling": "不靠堆数据 世界模型创新方法 竟让π0.5真机成功率最高提升12.8个百分点", "published_at": "2026-08-11T15:41:52+08:00", "updated_at": "2026-09-03T12:10:25+08:00"}
{"paper_id": "splatting-physical-scenes_2026_08", "slug": "splatting-physical-scenes", "title": "尺度不是模型给的，是机器人给的", "post_url": "https://haiguangboy.com/posts/splatting-physical-scenes", "asset_base_url": "https://static.haiguangboy.com/papers/splatting-physical-scenes", "tags": ["embodied_ai", "robotics", "state_prediction", "具身智能", "机器人学习", "3dgs", "real2sim"], "importance_score": 3.0, "ruling": "让机器人拿自己当尺子，重建误差 3.31mm，生成式模型差了近 5 倍", "published_at": "2026-08-10T16:07:47+08:00", "updated_at": "2026-09-03T12:15:15+08:00"}
{"paper_id": "video2robo_2026_08", "slug": "video2robo", "title": "闭的是运动学的环，不是物理的环", "post_url": "https://haiguangboy.com/posts/video2robo", "asset_base_url": "https://static.haiguangboy.com/papers/video2robo", "tags": ["embodied_ai", "action_generation", "robotics", "具身智能", "机器人学习", "3dgs", "数据生成"], "importance_score": 3.0, "ruling": "一段手机视频合成的数据，真机 62.5%，反超真机采集的 46.7%", "published_at": "2026-08-09T22:42:47+08:00", "updated_at": "2026-09-03T12:16:42+08:00"}
{"paper_id": "joyai-ra-dual-alignment_2026_08", "slug": "joyai-ra-dual-alignment", "title": "海量人类视频用不上、不同机器人的数据又拼不到一起——这篇论文把两个麻烦一起解决了", "post_url": "https://haiguangboy.com/posts/joyai-ra-dual-alignment", "asset_base_url": "https://static.haiguangboy.com/papers/joyai-ra-dual-alignment", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "异构数据"], "importance_score": 3.0, "ruling": "异构数据的问题不是量，是监督格式不统一", "published_at": "2026-08-09T13:29:52+08:00", "updated_at": "2026-09-03T12:10:40+08:00"}
{"paper_id": "omega-0_2026_08", "slug": "omega-0", "title": "未来预测越准，机器人反而越迟滞", "post_url": "https://haiguangboy.com/posts/omega-0", "asset_base_url": "https://static.haiguangboy.com/papers/omega-0", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "人形机器人", "世界模型"], "importance_score": 3.0, "ruling": "未来预测越准，机器人反而越迟滞", "published_at": "2026-08-09T08:43:52+08:00", "updated_at": "2026-09-03T12:12:08+08:00"}
{"paper_id": "transformer-transformer-codesign_2026_08", "slug": "transformer-transformer-codesign", "title": "机器人失败，可能不是策略不够大", "post_url": "https://haiguangboy.com/posts/transformer-transformer-codesign", "asset_base_url": "https://static.haiguangboy.com/papers/transformer-transformer-codesign", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "机器人", "本体设计", "世界模型"], "importance_score": 3.0, "ruling": "机器人失败，可能不是策略不够大", "published_at": "2026-08-08T12:56:40+08:00", "updated_at": "2026-09-03T12:16:32+08:00"}
{"paper_id": "rl-100_2026_08", "slug": "rl-100", "title": "RL 不该从零学，该做后训练", "post_url": "https://haiguangboy.com/posts/rl-100", "asset_base_url": "https://static.haiguangboy.com/papers/rl-100", "tags": ["embodied_ai", "action_generation", "robotics", "具身智能", "强化学习", "机器人策略"], "importance_score": 3.0, "ruling": "RL 不该从零学，该做后训练", "published_at": "2026-08-07T16:03:15+08:00", "updated_at": "2026-09-03T12:14:12+08:00"}
{"paper_id": "zprl_2026_08", "slug": "zprl", "title": "RL 该在哪一层介入", "post_url": "https://haiguangboy.com/posts/zprl", "asset_base_url": "https://static.haiguangboy.com/papers/zprl", "tags": ["embodied_ai", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "强化学习", "机器人策略"], "importance_score": 3.0, "ruling": "RL 该在哪一层介入", "published_at": "2026-08-07T07:52:06+08:00", "updated_at": "2026-09-03T12:17:26+08:00"}
{"paper_id": "dexora_2026_08", "slug": "dexora", "title": "双臂 + 双灵巧手 36 自由度的开源 VLA：灵巧任务 66.7%，比 GR00T N1 高 15 个点", "post_url": "https://haiguangboy.com/posts/dexora", "asset_base_url": "https://static.haiguangboy.com/papers/dexora", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "灵巧操作"], "importance_score": 3.0, "ruling": "遥操作数据的噪声得进损失函数", "published_at": "2026-08-05T18:19:52+08:00", "updated_at": "2026-09-03T12:08:32+08:00"}
{"paper_id": "mt3-thousand-tasks_2026_08", "slug": "mt3-thousand-tasks", "title": "1 天 1000 任务，靠的是归纳偏置", "post_url": "https://haiguangboy.com/posts/mt3-thousand-tasks", "asset_base_url": "https://static.haiguangboy.com/papers/mt3-thousand-tasks", "tags": ["embodied_ai", "action_generation", "robotics", "latent_state", "具身智能", "模仿学习", "数据效率"], "importance_score": 3.0, "ruling": "1 天 1000 任务，靠的是归纳偏置", "published_at": "2026-08-05T10:54:28+08:00", "updated_at": "2026-09-03T12:11:45+08:00"}
{"paper_id": "pegasus-editable-experience_2026_08", "slug": "pegasus-editable-experience", "title": "跨本体迁移该在经验层做", "post_url": "https://haiguangboy.com/posts/pegasus-editable-experience", "asset_base_url": "https://static.haiguangboy.com/papers/pegasus-editable-experience", "tags": ["embodied_ai", "robotics", "state_prediction", "具身智能", "跨本体迁移", "机器人数据"], "importance_score": 3.0, "ruling": "跨本体迁移该在经验层做", "published_at": "2026-08-04T16:45:50+08:00", "updated_at": "2026-09-03T12:12:51+08:00"}
{"paper_id": "tacwam_2026_08", "slug": "tacwam", "title": "触觉预测只该用来训练，不该给动作看", "post_url": "https://haiguangboy.com/posts/tacwam", "asset_base_url": "https://static.haiguangboy.com/papers/tacwam", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "触觉感知", "世界模型"], "importance_score": 3.0, "ruling": "触觉预测只该用来训练，不该给动作看", "published_at": "2026-08-03T18:18:19+08:00", "updated_at": "2026-09-03T12:16:11+08:00"}
{"paper_id": "archon-whole-body-intelligence_2026_07", "slug": "archon-whole-body-intelligence", "title": "李弘扬老师的全身智能高于GR的全身控制", "post_url": "https://haiguangboy.com/posts/archon-whole-body-intelligence", "asset_base_url": "https://static.haiguangboy.com/papers/archon-whole-body-intelligence", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "人形机器人", "全身智能"], "importance_score": 3.0, "ruling": "李弘扬老师的全身智能高于GR的全身控制", "published_at": "2026-08-02T06:51:40+08:00", "updated_at": "2026-09-03T12:08:09+08:00"}
{"paper_id": "dr-lfd_2026_08", "slug": "dr-lfd", "title": "50 条演示打 100%，别人 1000 条只有 45%——差别是它把「什么时候换招」这件事从策略里拿走了", "post_url": "https://haiguangboy.com/posts/dr-lfd", "asset_base_url": "https://static.haiguangboy.com/papers/dr-lfd", "tags": ["embodied_ai", "action_generation", "robotics", "state_prediction", "具身智能", "机器人学习", "tamp"], "importance_score": 3.0, "ruling": "切换逻辑不该让策略去学", "published_at": "2026-08-01T23:08:28+08:00", "updated_at": "2026-09-03T12:08:50+08:00"}
{"paper_id": "gemini-robotics-2_2026_07", "slug": "gemini-robotics-2", "title": "被忽略的非共识是全身统一控制", "post_url": "https://haiguangboy.com/posts/gemini-robotics-2", "asset_base_url": "https://static.haiguangboy.com/papers/gemini-robotics-2", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "具身智能", "机器人", "geminirobotics"], "importance_score": 3.0, "ruling": "被忽略的非共识是全身统一控制", "published_at": "2026-07-31T12:45:53+08:00", "updated_at": "2026-09-03T12:09:42+08:00"}
{"paper_id": "leapbot-wa_2026_07", "slug": "leapbot-wa", "title": "预测式特征不能直接喂扩散模型", "post_url": "https://haiguangboy.com/posts/leapbot-wa", "asset_base_url": "https://static.haiguangboy.com/papers/leapbot-wa", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "世界模型", "jepa"], "importance_score": 3.0, "ruling": "预测式特征不能直接喂扩散模型", "published_at": "2026-07-31T06:32:33+08:00", "updated_at": "2026-09-03T12:11:11+08:00"}
{"paper_id": "tactile-glove-pose-artifacts_2026_07", "slug": "tactile-glove-pose-artifacts", "title": "手套触觉读数里混着手的姿态", "post_url": "https://haiguangboy.com/posts/tactile-glove-pose-artifacts", "asset_base_url": "https://static.haiguangboy.com/papers/tactile-glove-pose-artifacts", "tags": ["embodied_ai", "world_model", "vla", "robotics", "state_prediction", "具身智能", "触觉感知", "机器人学习"], "importance_score": 3.0, "ruling": "手套触觉读数里混着手的姿态", "published_at": "2026-07-29T22:27:41+08:00", "updated_at": "2026-09-03T12:15:59+08:00"}
{"paper_id": "cd-lam_2026_07", "slug": "cd-lam", "title": "世界模型不听话，是潜在动作被污染了", "post_url": "https://haiguangboy.com/posts/cd-lam", "asset_base_url": "https://static.haiguangboy.com/papers/cd-lam", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "具身智能", "世界模型", "机器人学习"], "importance_score": 3.0, "ruling": "世界模型不听话，是潜在动作被污染了", "published_at": "2026-07-29T20:37:27+08:00", "updated_at": "2026-09-03T12:08:21+08:00"}
{"paper_id": "pi-rl-chelsea-finn-talk_2026_07", "slug": "pi-rl-chelsea-finn-talk", "title": "机器人 RL 的瓶颈是采样不是算法", "post_url": "https://haiguangboy.com/posts/pi-rl-chelsea-finn-talk", "asset_base_url": "https://static.haiguangboy.com/papers/pi-rl-chelsea-finn-talk", "tags": ["embodied_ai", "world_model", "robotics", "具身智能", "机器人学习", "强化学习"], "importance_score": 3.0, "ruling": "机器人 RL 的瓶颈是采样不是算法", "published_at": "2026-07-29T16:59:36+08:00", "updated_at": "2026-09-03T12:13:02+08:00"}
{"paper_id": "n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens_2026_07", "slug": "n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens", "title": "触觉当预测目标而非观测输入", "post_url": "https://haiguangboy.com/posts/n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens", "asset_base_url": "https://static.haiguangboy.com/papers/n_0_vtla_scaling_vision_tactile_language_action_model_with_latent_tactile_tokens", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "机器人学习", "触觉感知"], "importance_score": 3.0, "ruling": "触觉当预测目标而非观测输入", "published_at": "2026-07-29T08:55:47+08:00", "updated_at": "2026-09-03T12:11:56+08:00"}
{"paper_id": "robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode_2026_07", "slug": "robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode", "title": "不是\"要不要世界模型\"，是\"它吐出来的东西有没有结构\"", "post_url": "https://haiguangboy.com/posts/robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode", "asset_base_url": "https://static.haiguangboy.com/papers/robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "世界模型", "中间表征"], "importance_score": 3.0, "ruling": "不是\"要不要世界模型\"，是\"它吐出来的东西有没有结构\"", "published_at": "2026-07-24T11:40:57+08:00", "updated_at": "2026-09-03T12:14:29+08:00"}
{"paper_id": "liangwenfeng-world-model_2026_07", "slug": "liangwenfeng-world-model", "title": "这句话缺了一个前提：你的数据管道通没通", "post_url": "https://haiguangboy.com/posts/liangwenfeng-world-model", "asset_base_url": "https://static.haiguangboy.com/papers/liangwenfeng-world-model", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "具身智能", "世界模型", "deepseek", "技术路线"], "importance_score": 3.0, "ruling": "这句话缺了一个前提：你的数据管道通没通", "published_at": "2026-07-23T15:00:13+08:00", "updated_at": "2026-09-03T12:11:24+08:00"}
{"paper_id": "real_time_semg_based_telecontrol_of_an_assistive_robotic_arm_using_a_1d_convolut_2026_07", "slug": "real_time_semg_based_telecontrol_of_an_assistive_robotic_arm_using_a_1d_convolut", "title": "神经信号：它的价值和它的脆弱，来自同一处", "post_url": "https://haiguangboy.com/posts/real_time_semg_based_telecontrol_of_an_assistive_robotic_arm_using_a_1d_convolut", "asset_base_url": "https://static.haiguangboy.com/papers/real_time_semg_based_telecontrol_of_an_assistive_robotic_arm_using_a_1d_convolut", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "肌电", "数据采集", "机器人学习"], "importance_score": 3.0, "ruling": "神经信号：它的价值和它的脆弱，来自同一处", "published_at": "2026-07-23T10:54:07+08:00", "updated_at": "2026-09-03T12:14:02+08:00"}
{"paper_id": "sudu-third-position_2026_07", "slug": "sudu-third-position", "title": "苏度科技：横切一刀的第三种位置", "post_url": "https://haiguangboy.com/posts/sudu-third-position", "asset_base_url": "https://static.haiguangboy.com/papers/sudu-third-position", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "state_prediction", "具身智能", "机器人", "技术路线", "苏度科技"], "importance_score": 3.0, "ruling": "苏度科技：横切一刀的第三种位置", "published_at": "2026-07-23T09:17:33+08:00", "updated_at": "2026-09-03T12:15:25+08:00"}
{"paper_id": "latent_action_pretraining_through_world_modeling_2026_07", "slug": "latent_action_pretraining_through_world_modeling", "title": "LAWM：为什么动作标签会成为负担", "post_url": "https://haiguangboy.com/posts/latent_action_pretraining_through_world_modeling", "asset_base_url": "https://static.haiguangboy.com/papers/latent_action_pretraining_through_world_modeling", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "state_prediction", "具身智能", "世界模型", "预训练", "数据效率", "机器人学习"], "importance_score": 3.0, "ruling": "LAWM：为什么动作标签会成为负担", "published_at": "2026-07-22T10:41:01+08:00", "updated_at": "2026-09-03T12:10:55+08:00"}
{"paper_id": "qianxun-general-brain-data-loop_2026_07", "slug": "qianxun-general-brain-data-loop", "title": "千寻智能：通用大脑与数据闭环的双重赌注", "post_url": "https://haiguangboy.com/posts/qianxun-general-brain-data-loop", "asset_base_url": "https://static.haiguangboy.com/papers/qianxun-general-brain-data-loop", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "具身智能", "机器人数据", "通用机器人"], "importance_score": 3.0, "ruling": "千寻智能的数据闭环赌注", "published_at": "2026-07-21T22:35:35+08:00", "updated_at": "2026-09-03T12:29:13+08:00"}
{"paper_id": "t_rex_tactile_reactive_dexterous_manipulation_2026_07", "slug": "t_rex_tactile_reactive_dexterous_manipulation", "title": "T-Rex：触觉为何要单独建模", "post_url": "https://haiguangboy.com/posts/t_rex_tactile_reactive_dexterous_manipulation", "asset_base_url": "https://static.haiguangboy.com/papers/t_rex_tactile_reactive_dexterous_manipulation", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "触觉传感", "vla架构", "灵巧操作", "机器人学习"], "importance_score": 3.0, "ruling": "T-Rex：触觉为何要单独建模", "published_at": "2026-07-21T17:41:47+08:00", "updated_at": "2026-09-03T12:15:48+08:00"}
{"paper_id": "lingchu-optical_2026_07", "slug": "lingchu-optical", "title": "灵初智能：数据策略要看场景复杂度", "post_url": "https://haiguangboy.com/posts/lingchu-optical", "asset_base_url": "https://static.haiguangboy.com/papers/lingchu-optical", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "具身智能", "数据采集", "架构之争"], "importance_score": 3.0, "ruling": "灵初智能：数据策略要看场景复杂度", "published_at": "2026-07-18T19:32:44+08:00", "updated_at": "2026-09-03T12:11:33+08:00"}
{"paper_id": "sunday-act2_2026_07", "slug": "sunday-act2", "title": "Sunday的资源集中战略", "post_url": "https://haiguangboy.com/posts/sunday-act2", "asset_base_url": "https://static.haiguangboy.com/papers/sunday-act2", "tags": ["embodied_ai", "action_generation", "robotics", "具身智能", "机器人创业", "路线之争"], "importance_score": 3.0, "ruling": "Sunday的资源集中战略", "published_at": "2026-07-18T11:28:45+08:00", "updated_at": "2026-09-03T12:15:35+08:00"}
{"paper_id": "xsquare-robot_2026_06", "slug": "xsquare-robot", "title": "自变量：信念趋同不是证据趋同", "post_url": "https://haiguangboy.com/posts/xsquare-robot", "asset_base_url": "https://static.haiguangboy.com/papers/xsquare-robot", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "具身智能", "机器人创业", "资本逻辑"], "importance_score": 3.0, "ruling": "自变量：信念趋同不是证据趋同", "published_at": "2026-07-18T09:00:29+08:00", "updated_at": "2026-09-03T12:17:03+08:00"}
{"paper_id": "xuhuazhe-poke_2026_03", "slug": "xuhuazhe-poke", "title": "许华哲的组织下注", "post_url": "https://haiguangboy.com/posts/xuhuazhe-poke", "asset_base_url": "https://static.haiguangboy.com/papers/xuhuazhe-poke", "tags": ["embodied_ai", "vla", "robotics", "具身智能", "机器人创业", "路线之争"], "importance_score": 3.0, "ruling": "许华哲的组织下注", "published_at": "2026-07-18T08:59:52+08:00", "updated_at": "2026-09-03T12:17:15+08:00"}
{"paper_id": "schrodinger-navigator_2026_06", "slug": "schrodinger-navigator", "title": "导航机器人学会先想再走，遮挡难题有解了？", "post_url": "https://haiguangboy.com/posts/schrodinger-navigator", "asset_base_url": "https://static.haiguangboy.com/papers/schrodinger-navigator", "tags": ["embodied_ai", "world_model", "robotics", "state_prediction", "具身智能", "机器人导航", "世界模型", "零样本", "四足机器人"], "importance_score": 3.0, "ruling": "零样本导航：Schrödinger's Navigator", "published_at": "2026-07-18T08:59:17+08:00", "updated_at": "2026-09-03T12:14:40+08:00"}
{"paper_id": "psi-zero_2026_07", "slug": "psi-zero", "title": "堆10倍数据还输40%--Psi-Zero说混训人形机器人根本就是错的", "post_url": "https://haiguangboy.com/posts/psi-zero", "asset_base_url": "https://static.haiguangboy.com/papers/psi-zero", "tags": ["embodied_ai", "vla", "action_generation", "robotics", "具身智能", "人形机器人", "机器人学习"], "importance_score": 3.0, "ruling": "少量机器人数据下的策略学习", "published_at": "2026-07-07T19:22:32+08:00", "updated_at": "2026-09-03T12:13:37+08:00"}
{"paper_id": "fast-wam_2026_07", "slug": "fast-wam", "title": "推理时不用想象未来，机器人照样跑到91.8%！Fast-WAM拆穿了WAM的核心假设", "post_url": "https://haiguangboy.com/posts/fast-wam", "asset_base_url": "https://static.haiguangboy.com/papers/fast-wam", "tags": ["embodied_ai", "world_model", "action_generation", "robotics", "具身智能", "世界模型", "fastwam", "机器人学习"], "importance_score": 3.0, "ruling": "训练视频目标，比测试时想象未来更关键", "published_at": "2026-07-07T15:32:00+08:00", "updated_at": "2026-09-03T12:09:06+08:00"}
{"paper_id": "orca_2026_07", "slug": "orca", "title": "π0.5抓勺子失败就原地哆嗦,Orca却靠刷视频悟出的物理直觉更进一步", "post_url": "https://haiguangboy.com/posts/orca", "asset_base_url": "https://static.haiguangboy.com/papers/orca", "tags": ["embodied_ai", "world_model", "vla", "action_generation", "robotics", "latent_state", "state_prediction", "具身智能", "世界模型"], "importance_score": 3.0, "ruling": "世界模型的关键是可读出的状态", "published_at": "2026-07-06T23:19:24+08:00", "updated_at": "2026-09-03T12:12:34+08:00"}