TouchWorld: A Predictive and Reactive Tactile Foundation Model for Dexterous Manipulation

Jianyi Zhou^*, Feiyang Hong^*, Yunhao Li^*, Yicheng Zhao, Yongjue Cen, Zirui Liu, Jiakang Huang, Zirui Chen, Ruiyang Zhang, Weizhuo Zhu, Xuhua Song, Shuo Yang^†

Harbin Institute of Technology, Shenzhen; PHANES AI   (^* Equal contribution; ^† Corresponding author)

Project Page / Paper

• A predictive-and-reactive tactile foundation model that combines contact-aware tactile subgoal prediction with high-frequency residual refinement for robust dexterous manipulation.

TouchAnything: A Dataset and Framework for Bimanual Tactile Estimation from Egocentric Video

Jianyi Zhou, Ziteng Gao, Feiyang Hong, Zirui Liu, Guannan Zhang, Weisheng Dai, Ruichen Zhen, Chuqiao Lyu, Haotian Wu, Yinian Mao, Xushi Wang, Yuxiang Jiang, Shuo Yang

Harbin Institute of Technology, Shenzhen; Meituan Academy of Robotics

Project Page

• The first large-scale multi-view tactile dataset for egocentric hand-object interaction with bimanual 3D hand pose annotations and dense continuous pressure maps.

KFC-Agent: Kernel Fault Corrector-Agent for Automated Linux Kernel Crash Repair

Qin Yuhuai, Jianyi Zhou, Yue Liang

Harbin Institute of Technology, Shenzhen

Code Repository

• A multi-turn agent that explores crash reports, kernel source code, and execution feedback through an interactive Docker/SWE-ReX environment instead of generating a patch from a single long context.
• Combines extensible tools with trace-based call-graph retrieval, a Planner built from historical crash reports and developer discussions, and a Memory module for reusable debugging experience.
• On kbench with DeepSeek-V3 and four patches, KFC-Agent reaches 60.22% execution success, compared with 6.45% for kGym; its per-instance cost is $0.28.

Inject Once Survive Later: Backdooring Vision-Language-Action Models to Persist Through Downstream Fine-tuning

Jianyi Zhou, Yujie Wei, Ruichen Zhen, Bo Zhao, Xiaobo Xia, Rui Shao, Xiu Su, Shuo Yang

Harbin Institute of Technology, Shenzhen; Meituan Academy of Robotics; Shanghai Jiaotong University; National University of Singapore; Central South University

Paper

• Research on backdoor attacks and security vulnerabilities in vision-language-action models for embodied AI systems.

ConLA: Contrastive Latent Action Learning from Human Videos for Robotic Manipulation

Weisheng Dai, Kai Lan, Jianyi Zhou, Bo Zhao, Xiu Su, Junwen Tong, Weili Guan, Shuo Yang

Harbin Institute of Technology, Shenzhen; ZTE Corporation; Shanghai Jiao Tong University; Central South University

Paper

• Contrastive latent action learning framework for robotic manipulation from human demonstration videos.