https://iconlab.negarmehr.com/POLICEd-RL/ https://danijar.com/project/daydreamer/ https://github.com/JiangWenPL/FisherRF https://github.com/AlbertTan404/RoLD Diffusion Meets DAgger: Supercharging Eye-in-hand Imitation Learning MaskViT: Masked Visual Pre-Training for Video Prediction
- https://navila-bot.github.io/ NaVILA,Nvidia做得一个用 VLA 做 Navigation 的工作
https://shihao1895.github.io/MemoryVLA/ 加了 Memory 的 VLA
CTRL-WORLD: A CONTROLLABLE GENERATIVE WORLD MODEL FOR ROBOT MANIPULATION https://arxiv.org/pdf/2510.10125 可以理解和生成多视角的 World Model。
-
ActiveGlasses: Learning Manipulation with Active Vision from Ego-centric Human Demonstrat. 用 Diffusion Policy 来模仿头部 camera 的轨迹
-
EgoAVFlow: Robot Policy Learning with Active Vision from Human Egocentric Videos via 3D Flow.
-
https://github.com/Junyi42/LoGeR 为 VGGT 添加长程 memory。
-
https://arxiv.org/pdf/2605.06270 Spark3R,在 VGGT 基础上直接做 token 压缩,从而达到20倍以上加速。
-
https://www.liuisabella.com/LoHoManip/ Trace + VLA 做多步操作。
-
https://internrobotics.github.io/internvla-m1.github.io/ InternVLA-M1,(trace) conditioned vla 的比较开创性的工作
- [2026 ICLR] ST4VLA: Spatially Guided Training for Vision-Language-Action Models 看上去 InternVLA-M1 没有中,重新投了这个
-
https://www.micdz.cn/Traj2Action/ Traj2Action,同样 trace + vla,但是额外有专门的 trajectory expert 和 action expert
-
https://github.com/IRMVLab/MMTwin MMTwin,预测第一人称动态视角下,人手动作轨迹
- Novel Diffusion Models for Multimodal 3D Hand Trajectory Prediction
- Uni-Hand: Universal Hand Motion Forecasting in Egocentric Views
-
https://github.com/amap-cvlab/ABot-Manipulation ABot-Manipulation,阿里的 VLA 模型,看上去接入了 3D 信息,并且对 action space 进行了比较独特的设计。