Robust Stable Manipulation with Dexterous Hands under Changing External Forces

Dexterous hands operating in the real world routinely face time-varying external wrenches (e.g., tool reaction forces, intermittent environmental contacts, human–robot interaction, impacts, shifting loads). These disturbances can induce slip, loss of stable contact, and actuator saturation—leading to task failure and safety risks. Despite rapid progress in dexterous manipulation, robust stability under changing external forces remains a fundamental bottleneck for deploying dexterous hands outside carefully controlled settings.

Building on our previous work [1] on imitation-guided bimanual planning for stable manipulation under changing external forces—which demonstrated that stability-aware planning can be achieved via efficient grasp transitions within a hierarchical architecture integrating imitation learning and constrained optimization—we aim to extend the same stability-aware philosophy from gripper-based bimanual systems to dexterous, multi-finger hands. However, several key challenges must be addressed for this extension:

• Multi-contact state explosion: Moving from two grippers to a dexterous hand introduces many simultaneous contacts with continuously varying locations/normals, making the contact configuration and transition space far higher-dimensional than the grasp-transition abstraction used before.

• Force allocation under constraints: Dexterous hands require explicit multi-contact wrench/force distribution (including internal forces) that must satisfy friction limits and tight actuator/joint constraints; feasibility in a gripper setting does not directly translate to realizable stability for multi-finger actuation.

• Real-time robustness to uncertainty: Stable dexterous manipulation under changing external forces typically needs high-rate online adaptation and is highly sensitive to friction/contact uncertainty and compliance effects, which are less critical (or easier to manage) in gripper-based bimanual setups.

Project Goals

This project therefore extends [1] toward dexterous manipulation, targeting a practical and research-relevant capability: stable manipulation under changing external forces using a dexterous hand. The approach combines:

• Stability assessment: an online feasibility/risk estimator producing a continuous stability margin signal;
• Constrained control: a controller that generates actions while satisfying stability-related constraints (e.g., contact/friction/actuation limits); and
• Optimization-based reconfiguration: limited, executable changes in strategy/contact usage when margins degrade, with an explicit bias toward minimal switching.

Expected Outcome

This project builds on our previous published work and aims to produce a publishable outcome, targeting top robotics venues and journals such as IROS, ICRA, IEEE RA-L, and IEEE T-ASE.

Candidate Requirements

Time Commitment
We expect a commitment of 20+ hours per week. The project is task-driven: if you complete the planned milestones efficiently and with high quality, time spent will not be treated as a hard constraint.

Must-have

• C++ and Python programming skills
• Solid robotics fundamentals (control/optimization is a plus)

Nice-to-have

• Prior experience with contact-rich manipulation/control (e.g., grasping, multi-contact modeling)
• Familiarity with optimization-based control methods such as QP and MPC

Feel free to get in touch and discuss more details with me at [email protected]

References

[1] Cai, Kuanqi, Chunfeng Wang, Zeqi Li, Haowen Yao, Weinan Chen, Luis Figueredo, Aude Billard, and Arash Ajoudani. “Imitation-Guided Bimanual Planning for Stable Manipulation under Changing External Forces.” In 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 15932-15939. IEEE, 2025.