About Me

Bowen Jing^†, Mingxin Wang^†, Ruiyang Hao, Chenchen Ge, Hanwen Shen, Junjie He, Yang Cui, Yiming Hou, Weitao Zhou^*, Jiawei Wang, Minglei Li, Dandan Zhang, Ding Zhao, Houde Liu, Xiaofan Li, Si Liu, Ping Luo, Haibao Yu^*
† Equal contribution; ^* Corresponding author
arXiv preprint
arXiv / Code / Website

Introduces a safety-aware visuo-tactile benchmark for deformable object manipulation, evaluating both task success and physically safe interaction under contact-rich constraints.

AMAP CV Lab
arXiv preprint
arXiv / Code

Presents a unified world action model for mobility-and-manipulation tasks with temporal, action-space, and train-test alignment.

Mingxin Wang, Zhirun Yue, Renhao Lu, Yizhe Li, Zihan Wang, Guoping Pan, Kangkang Dong, Jun Cheng, Yi Cheng, Houde Liu^*
* Corresponding author
Accepted to IROS 2026
arXiv / Code

Proposes a phase-scheduled visual-force policy that integrates low-frequency diffusion planning and high-frequency residual force correction for contact-rich robotic manipulation tasks.

Push-Wiper: Toward General-Purpose Robotic Cleaning across Varied Stains and Surfaces with Segmented Pushing Trajectories

Renhao Lu^†, Mingxin Wang^†, Chenyang Cao, Yang Yang, Guoping Pan, Kangkang Dong, Yi Cheng, Houde Liu^*
† Equal contribution; ^* Corresponding author
Accepted to IROS 2026

Redefines viscous stain cleaning as an aggregation-post-processing task and enables zero-shot generalization for robotic cleaning on unseen stains and surfaces.

RRRR: Rapid Real-World Residual RL for Multi-Task VLA Adaptation

Yizhe Li^†, Zihan Wang^†, Zhe Han^†, Mingxin Wang, Guoping Pan, Yi Cheng, Yuheng Min, Xueqian Wang, Houde Liu^*
† Equal contribution; ^* Corresponding author
Under review

Presents a hybrid SFT+Residual RL framework to mitigate catastrophic forgetting and task interference in multi-task real-world VLA deployment.

Zhirun Yue, Mingxin Wang, Tianyi You, Jun Cheng, Houde Liu^*
* Corresponding author
Accepted to IJCNN 2026
PDF

Introduces a spatio-temporal augmented memory policy with a hierarchical memory pyramid for robust robotic manipulation in complex scenarios.

Ao Hong, Lehang Wang, Zhirun Yue, Mingxin Wang, Zihan Wang, Houde Liu^*
* Corresponding author
Under review

Proposes a training-free multi-agent pipeline that decomposes aspect sentiment triplet extraction into specialized extraction, reasoning, and consistency-checking stages for zero-shot deployment.

Zhirun Yue, Mingxin Wang, Ao Hong, Jun Cheng, Houde Liu^*
* Corresponding author
Under review

Formulates cluttered-scene grasping as pre-to-post decision reasoning, combining geometric-semantic evidence, VLM-based agentic reasoning, and contact-grasp refinement.

Mainstream WAM Evaluation Report

PDF

Benchmarks mainstream World Action Models under unified LIBERO and LIBERO-Plus protocols, analyzing standard-task performance, OOD generalization, inference efficiency, 3D capability, and failure taxonomies.

AGIBOT WORLD CHALLENGE @ ICRA 2026 World Model Track

Ranked Top 10

The challenge evaluates world models in terms of Visual Quality, Action Following, and Scene Consistency. We developed a WAN2.1-based action-conditioned world model for predicting future world dynamics.