Yongpeng Jiang

We introduce a tactile-driven model predictive controller for adaptive dexterous grasp execution, coordinating arm-hand motions and multi-contact forces across approaching and grasping phases to improve grasp success under uncertainty while reducing undesired in-hand object motion.

We introduce HUGS, a human-prior-guided framework for unified dexterous grasp synthesis across object scales and contact modes, enabling large-scale grasp generation and real-world autonomous selection from two-finger to bimanual grasps.

We introduce an arm-aware dexterous grasp generation framework that reuses pretrained hand-centric diffusion models while incorporating arm and environment constraints at inference time, enabling efficient feasible grasp synthesis in highly constrained real-world scenarios.

We present a comprehensive real-world comparative study of kinematic retargeting objectives for dexterous manipulation, revealing the significance of key objective terms through ablations and providing practical insights for more effective human-to-robot hand motion transfer.

We introduce a practical low-cost framework for in-hand cube reorientation, combining RL-trained motion primitives with a high-level switching policy to achieve long-duration continuous manipulation using only a LEAP Hand and a single RGB camera.

We introduce a tactile-incorporated bi-level framework for robust model-based dexterous in-hand manipulation, combining contact-implicit MPC for real-time planning with tactile-reactive force-motion control to handle modeling errors, disturbances, and diverse real-world tasks.

We introduce a practical kinematic trajectory optimization approach for in-grasp object movement, enabling dexterous fingers to achieve precise, large-range in-hand manipulation while maintaining stable grasps, without pretraining or object geometry information.

We introduce a unified human-intention-aware interaction control framework for ultrasound scanning robots, enabling smooth and compliant responses to intended interventions and unintended collisions while maintaining safe and efficient scanning performance.

We introduce a model-based framework for long-horizon dexterous in-hand manipulation, using contact-implicit MPC to generate real-time contact plans without predefined contact sequences, separate planning, or pretraining, enabling robust and generalizable object rotation.

We introduce a contact-aware non-prehensile manipulation framework for cluttered object retrieval, combining improved RRT planning with MPC control to enable a rod-like pusher to retrieve target objects through obstacle removal, avoidance, and contact-face switching.

We introduce a complementary human-robot collaboration framework that decouples robot task execution and human null-space intervention, enabling safe expert guidance under unforeseen changes while preserving task efficiency through vision-based adaptive control and DMP-based learning.