2026-02-12
-
YOR Your Own Mobile Manipulator for Generalizable Robotics (arXiv:2602.11150)
召回要点：低成本开源双臂移动操作平台，强调全身协同能力与可复现实验生态。 -
APEX Learning Adaptive High-Platform Traversal for Humanoid Robots (arXiv:2602.11143)
召回要点：以 ratchet progress reward + 多技能蒸馏实现超腿长高台安全跨越。 -
Data-Efficient Hierarchical Goal-Conditioned Reinforcement Learning via Normalizing Flows (arXiv:2602.11142)
召回要点：在层级 GCRL 中以 normalizing flow 替代高斯策略，强化多模态与低数据效率。 -
RISE Self-Improving Robot Policy with Compositional World Model (arXiv:2602.11075)
召回要点：组合世界模型做 imagination RL，显著减少真机交互依赖并提升动态操作鲁棒性。 -
ST4VLA Spatially Guided Training for Vision-Language-Action Models (arXiv:2602.10109)
召回要点：用空间监督双阶段训练避免 VLA 在动作学习中丢失几何 grounding。 -
EgoHumanoid Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration (arXiv:2602.10106)
召回要点：引入机器人无关第一视角示教并对齐视角与动作，缓解人形数据瓶颈。 -
VLA-JEPA Enhancing Vision-Language-Action Model with Latent World Model (arXiv:2602.10098)
召回要点：以 JEPA 潜空间预测替代像素重建，提升 action-relevant 表征鲁棒性。 -
BagelVLA Enhancing Long-Horizon Manipulation via Interleaved Vision-Language-Action Generation (arXiv:2602.09849)
召回要点：交错式语言推理、视觉前瞻与动作生成，提高长时程任务恢复力。 -
Rethinking Visual-Language-Action Model Scaling Alignment, Mixture, and Regularization (arXiv:2602.09722)
召回要点：证明 VLA 扩展关键在对齐与混合策略，盲目加数据会触发负迁移。 -
Sample-Efficient Real-World Dexterous Policy Fine-Tuning via Action-Chunked Critics and Normalizing Flows (arXiv:2602.09580)
召回要点：flow policy + chunk critic 在真机低预算下实现更稳健灵巧微调。 -
Certified Gradient-Based Contact-Rich Manipulation via Smoothing-Error Reachable Tubes (arXiv:2602.09368)
召回要点：将平滑接触误差显式包进 reachable tube，给梯度接触控制提供可认证安全界。 -
Agile asymmetric multi-legged locomotion contact planning via geometric mechanics and spin model duality (arXiv:2602.09123)
召回要点：几何力学结合 spin 对偶搜索，挖掘非对称步态的速度与容错优势。
2026-02-11
-
Mimic Intent, Not Just Trajectories (arXiv:2602.08602)
痛点与看点：模仿学习把意图和执行混在一起导致 OOD 易崩；这篇用多尺度动作表示显式解耦 intent/execution，迁移与鲁棒性更强。 -
Differentiate-and-Inject Enhancing VLAs via Functional Differentiation Induced by In-Parameter Structural Reasoning (arXiv:2602.07541)
痛点与看点：长任务中计划与控制纠缠；这篇把任务结构注入参数空间而非只靠提示词分解。 -
Self-Supervised Bootstrapping of Action-Predictive Embodied Reasoning (arXiv:2602.08167)
痛点与看点：推理很长但对动作未必有用；这篇自监督筛出“对动作预测有信息增益”的 reasoning。 -
Why Look at It at All Vision-Free Multifingered Blind Grasping Using Uniaxial Fingertip Force Sensing (arXiv:2602.07326)
痛点与看点：灵巧抓取常依赖昂贵视觉/触觉；这篇用极简传感（单轴力+本体感觉）实现高成功率盲抓。 -
TwinRL-VLA Digital Twin-Driven Reinforcement Learning for Real-World Robotic Manipulation (arXiv:2602.09023)
痛点与看点：真机在线 RL 在 OOD 区域学习效率低；这篇核心是用 twin 做状态分布重加权，显著提升样本效率。 -
DexFormer Cross-Embodied Dexterous Manipulation via History-Conditioned Transformer (arXiv:2602.08278)
痛点与看点：换手型几乎要重训；这篇用共享动作语义空间+时序建模，提升跨本体 zero-shot 泛化。 -
RLinf-USER A Unified and Extensible System for Real-World Online Policy Learning in Embodied AI (arXiv:2602.07837)
痛点与看点：真机在线学习常卡在系统瓶颈；这篇偏基础设施，强调调度、通信和异步训练闭环。 -
StreamVLA Breaking the Reason-Act Cycle via Completion-State Gating (arXiv:2602.01100)
痛点与看点：每步重推理导致高延迟和目标漂移；这篇通过 completion-state gating 实现“该想时想，该控时控”。 -
Learning Human-Like Badminton Skills for Humanoid Robots (arXiv:2602.08370)
痛点与看点：人形高动态全身协调难；这篇展示从模仿到交互打击的渐进式学习流程。 -
UniPlan Vision-Language Task Planning for Mobile Manipulation with Unified PDDL Formulation (arXiv:2602.08537)
痛点与看点：移动+操作任务规划碎片化；这篇统一到 PDDL 表达并连接视觉状态落地。 -
Leveraging Adaptive Group Negotiation for Heterogeneous Multi-Robot Collaboration with Large Language Models CLiMRS (arXiv:2602.06967)
痛点与看点：多机协作决策沟通成本高；这篇用多代理协商机制提升复杂协作效率。 -
SceneSmith Agentic Generation of Simulation-Ready Indoor Scenes (arXiv:2602.09153)
痛点与看点：机器人训练缺可仿真高密度场景数据；这篇以多 VLM agent 协作生成 simulation-ready 场景。 -
Agile asymmetric multi-legged locomotion contact planning via geometric mechanics and spin model duality (arXiv:2602.09123)
痛点与看点：多足接触规划组合爆炸；这篇用几何力学+统计物理对偶给出非对称步态新解。 -
Legs Over Arms On the Predictive Value of Lower-Body Pose for Human Trajectory Prediction from Egocentric Robot Perception (arXiv:2602.09076)
痛点与看点：社交导航里人体特征选取不清晰；这篇发现下肢 3D 姿态对轨迹预测贡献更大。