贾金豆 | 南洋理工大学博士后

你好，我是贾金豆

我目前是南洋理工大学（NTU）MARS 实验室的博士后，合作导师为杨剑飞教授。2025 年博士毕业于北京航空航天大学沈元荣誉学院，师从郭雷院士、余翔教授和郭克信教授。

我的研究方向是可泛化的机器人学习：将世界模型、生成式策略和不确定性预测与物理先验相结合，让机器人能够从训练环境可靠地迁移到从未见过的新环境中。

Scholar Github ResearchGate LinkedIn 小红书

新闻动态

2026年9月：🎉 GRIPPER 被 T-RO 接收！
2026年9月：🎉 Feedback world model 和 FLASH 被 2026 NeurIPS 接收！
2026年9月：🎉 ADAPT 被 2026 CoRL 接收！
2026年8月：🎉 PhyFilter 被 Nature npj Robotics 接收！
2026年8月：🎉 我们的综述《面向机器人学习的世界模型》被 IJRR 接收！
2026年5月：⭐ 发布 MARS policy：只在真正需要时引入多模态。
2026年4月：🎉 A2A policy 被 2026 RSS 接收！
2026年2月：🎉 两篇论文被 2026 ICRA 接收！
2025年11月：🚀 加入南洋理工大学，开始博士后工作！
2025年7月：🎉 FORESEER 被 IJRR 接收！
2025年6月：😊 顺利通过博士学位论文答辩！
2025年2月：🎉 Feedback neural network 入选 2025 ICLR 口头报告（Oral）（前 1.8%）！
2024年7月：🎉 我们的论文 TRACE 获得 IEEE ICCA 2024 最佳学生论文奖！
2023年10月：🎉 EVOLVER 被 T-RO 接收！

机器人平台

这些年我研发和使用过的机器人硬件平台。

论文

以下均为本人第一作者、共同第一作者^*或通讯作者^†的论文。完整列表见 Google Scholar →

CoRL'26

ADAPT: Analytical disturbance-aware policy training for humanoid locomotion

Bofan Lyu^*, Jindou Jia^*, Kuangji Zuo, Yanshuo Lu, Shijia Han, Gen Li, Boyu Ma, Jingliang Li, Geng Li, Jianfei Yang

Conference on Robot Learning (CoRL), 2026

We present ADAPT, an analytical disturbance-aware framework for humanoid locomotion that infers joint-level external disturbances from robot dynamics and proprioception alone, without specialized sensors, and feeds them back into policy training for robustness to pushes, payloads, and impacts.

项目主页 PDF arXiv

★ 精选

arXiv'26

Δ-Flow matching for robot foundation models

Kuangji Zuo, Tuo An, Xinying Guo, Gen Li, Mengfei Zhao, Jiaqi Bai, Bofan Lyu, Zhuoyuan Yu, Jinghan Yang, Jiayi Li, Jindou Jia^†, Jianfei Yang^†

arXiv, 2026

Scaling A2A to robot foundation models: Δ-flow starts flow matching from recent history instead of Gaussian noise, enabling single-step inference, faster training, and smoother actions.

项目主页 PDF arXiv

arXiv'26

MOSS: Mode-oriented source shaping for robot flow matching policies

Zirun Ma^*, Jindou Jia^*, Tuo An, Jiaqi Bai, Jiahua Dong, Jianfei Yang

arXiv, 2026

MOSS makes behavior-mode selection controllable in robot flow matching policies by shaping mode-specific source distributions, separating modes only where the situation is truly multimodal.

项目主页 PDF arXiv

arXiv'26

TUCO: Curating simulation demonstrations for sim-to-real robot policy co-training

Ning Zhu, Mengfei Zhao, Yikai Tang, Zhangyujie Sun, Peihao Li, Dongyue Ni, Jindou Jia^†, Jianfei Yang^†

arXiv, 2026

TUCO curates simulation demonstrations for sim-to-real policy co-training, using influence functions to score each demonstration's trajectory utility and set coverage.

PDF arXiv

arXiv'26

APEX: Adaptive policy execution for precise manipulation

Mengfei Zhao, Chenxi Jiang, Tuo An, Jindou Jia^†, Jianfei Yang^†

arXiv, 2026

APEX is a plug-and-play framework that bridges the execution gap of learned manipulation policies by reconstructing dynamically feasible references from policy outputs and adapting them online via state feedback.

项目主页 PDF arXiv

★ 精选

arXiv'26

MARS policy: Multimodality only when it matters

Jindou Jia^*, Tuo An^*, Yuxuan Hu^*, Gen Li, Jingliang Li, Bohan Hou, Xiangyu Chen, Jiaqi Bai, Bofan Lyu, Jianfei Yang

arXiv, 2026

We present MARS, a generative policy that adaptively invokes stochasticity only when behavioral branching truly matters and reverts to deterministic learning otherwise, yielding 16.67% higher success and 83.20% lower inference latency in real-world tests.

项目主页 PDF arXiv 视频

NeurIPS'26

Feedback world model enables precise guidance of diffusion policy

Tuo An^*, Jindou Jia^*, Gen Li, Jingliang Li, Chuhao Zhou, Pengfei Liu, Bofan Lyu, Jiaqi Bai, Xinying Guo, Geng Li, Jianfei Yang

Conference on Neural Information Processing Systems (NeurIPS), 2026

We close the prediction–observation loop in world models with a lightweight feedback state that corrects future predictions online, paired with action-aware guidance to emphasize controllable dimensions, reducing prediction error by up to 76.4% and improving OOD success by 30%.

项目主页 PDF arXiv 代码视频

NeurIPS'26

FLASH: Efficient visuomotor policy via sparse sampling

Jiaqi Bai^*, Jindou Jia^*, Yuxuan Hu, Gen Li, Xiangyu Chen, Tuo An, Kuangji Zuo, Jianfei Yang

Conference on Neural Information Processing Systems (NeurIPS), 2026

We present FLASH, an efficient visuomotor policy that replaces iterative denoising with continuous Legendre polynomial trajectories, enabling single-step inference for real-time robot control.

项目主页 PDF arXiv 代码视频

★ 精选

npj Robotics'26

Physics filtering favors the generalization of robot learning

Jindou Jia^*, Shixuan Han^*, Meng Wang^*, Gen Li, Zihan Yang, Sicheng Zhou, Kexin Guo^†, Jianfei Yang^†, Xiang Yu, Wei Wang, Lei Guo

npj Robotics (Nature), 2026

We propose PhyFilter, which enhances both generalization and interpretability of robot learning. PhyFilter operates by correcting learning outcomes according to readily accessible physical differential structure and real-time state feedback.

链接项目主页 PDF arXiv 代码

IJRR'26

World model for robot learning: A comprehensive survey

Bohan Hou^*, Gen Li^*, Jindou Jia^*, Tuo An^*, Xinying Guo^*, Sicong Leng, Haoran Geng, Yanjie Ze, Tatsuya Harada, Philip Torr, Oier Mees, Marc Pollefeys, Zhuang Liu, Jiajun Wu, Pieter Abbeel, Jitendra Malik, Yilun Du, Jianfei Yang

The International Journal of Robotics Research (IJRR), 2026

A comprehensive survey of world models for robot learning, covering world models as policies and as simulators across recent literature.

链接项目主页 PDF arXiv 代码

★ 精选

RSS'26

Action-to-action flow matching

Jindou Jia^*, Gen Li^*, Xiangyu Chen, Tuo An, Yuxuan Hu, Jingliang Li, Xinying Guo, Jianfei Yang

Robotics: Science and Systems (RSS), 2026

We introduces A2A, an efficient generative paradigm that replaces noise-based initialization with action-to-action transport.

项目主页 PDF arXiv 视频

ICRA'26

Learning-based observer for coupled disturbance

Jindou Jia^*, Meng Wang^*, Zihan Yang, Bin Yang, Yuhang Liu, Kexin Guo, Xiang Yu

IEEE International Conference on Robotics and Automation (ICRA), 2026

We present a learning-based observer, enabling accurate prediction of the coupled disturbance consisting of internal uncertainties and external disturbances.

PDF arXiv

arXiv'26

Optimizing control-friendly trajectories with self-supervised residual learning

Kexin Guo, Zihan Yang, Yuhang Liu, Jindou Jia^†, Xiang Yu

arXiv, 2026

The presented trajectory optimizer outputs trajectories that are friendly to the following control level.

PDF arXiv

★ 精选

IJRR'25

FORESEER: Recognize and utilize uncertainties by integrating data-based learning and symbolic feedback

Jindou Jia, Kexin Guo, Yuyang Wang, Sicheng Zhou, Jiayi Zhang, Yuhang Liu, Xiang Yu, Yang Shi, Lei Guo

The International Journal of Robotics Research (IJRR), 2025

We present two converged uncertainty prediction frameworks, enabling accurate prediction of two general kinds of uncertainties, respectively.

链接项目主页视频

★ 精选

ICLR'25

Feedback favors the generalization of neural ODEs

Jindou Jia^*, Zihan Yang^*, Meng Wang, Kexin Guo, Jianfei Yang, Xiang Yu, Lei Guo

International Conference on Learning Representations (ICLR), 2025

Oral Presentation

We present feedback neural networks, showing that a feedback loop can flexibly correct the learned latent dynamics of neural ordinary differential equations (neural ODEs), leading to a prominent generalization improvement.

链接项目主页 PDF arXiv 代码

★ 精选

T-RO'24

EVOLVER: Online learning and prediction of disturbances for robot control

Jindou Jia, Wenyu Zhang, Kexin Guo, Jianliang Wang, Xiang Yu, Yang Shi, Lei Guo

IEEE Transactions on Robotics (T-RO), 2024

We present a framework, namely EVOLVER, to mimic the bio-behavior for robotics to achieve rapid transient reaction ability and high precision steady-state performance simultaneously.

链接项目主页代码视频

★ 精选

R-AL'22

Accurate high-maneuvering trajectory tracking for quadrotors: A drag utilization method

Jindou Jia, Kexin Guo, Xiang Yu, Weihua Zhao, Lei Guo

IEEE Robotics and Automation Letters (R-AL), 2022

Different from standard approaches that achieve precise tracking by feedforward compensating the estimated drag, this work presents a scheme to appropriately utilize drag.

链接视频

T-AES'22

Agile flight control under multiple disturbances for quadrotor: Algorithms and evaluation

Jindou Jia, Kexin Guo, Xiang Yu, Lei Guo, Lihua Xie

IEEE Transactions on Aerospace and Electronic Systems (T-AES), 2022

A scheme of anti-disturbance agile flight control is developed for a maneuverable quadrotor unmanned aerial vehicle, subject to the aerodynamic drag, dynamic shift of center of gravity, and motor dynamics.

链接视频

CEP'20

Multiple observers-based anti-disturbance control for a quadrotor UAV against payload and wind disturbance

Kexin Guo^*, Jindou Jia^*, Xiang Yu, Lei Guo, Lihua Xie

Control Engineering Practice, 2020

This paper presents a multiple observers-based anti-disturbance control scheme against multiple disturbances for a quadrotor unmanned aerial vehicle.

链接视频

ICUAS'22

Flight control for quadrotor safety in the presence of CoG shift and loss of motor efficiency

Jindou Jia, Kexin Guo, Chenliang Wang, Ning Shen, Wei Jia, Xiang Yu

International Conference on Unmanned Aircraft Systems (ICUAS), 2022

Oral Presentation

A safety control strategy based on a novel nonlinear disturbance observer and geometric control is developed for a quadrotor unmanned aerial vehicle, subject to the disturbances caused by center-of-gravity shift and loss of motor efficiency.

链接视频

CAC'19

Dual-disturbance observers-based control of UAV subject to internal and external disturbances

Kexin Guo^*, Jindou Jia^*, Xiang Yu, Lei Guo

China Automation Conference, 2019

First Prize of Outstanding Paper

This paper presents an embedded micro loop to enhance the anti-disturbance performance for unmanned aerial vehicles.

链接视频

报告与媒体

2026年10月特邀报告

Classical Control and Planning Meet Modern Learning

南京大学（NJU）· Shangke Lyu 教授邀请

2026年8月访谈

Robopapers Podcast

受邀访谈：A2A policy

2026年6月特邀报告

具身智能机器人研讨会

高效机器人策略学习 · 新加坡 SIMTech & ARTC

2025年3月特邀报告

TechBeat

特邀技术报告：反馈神经常微分方程（Feedback Neural ODE）

学术服务

副编辑（Associate Editor）

IEEE Robotics and Automation Letters (RA-L)

审稿人

T-RO、IJRR、RA-L、ICRA 等机器人领域期刊与会议