Motivation
Compliance is important in robotic manipulation [1]: it trades off position accuracy against the forces the robot applies on contact. Most Imitation Learning frameworks focus solely on position, without controlling contact force [2]. This approach can be limiting for dexterous manipulation, where contact forces play an important role [3]. We would like a control policy that decides both where to move, and how compliant to be, with compliance that can be tuned externally.
Outline
This project aims to build an imitation learning framework for a 20-DoF dexterous hand (Wuji Hand [5]) and a 7-DoF robotic arm, commanded only in position. We implement an admittance controller and collect demonstrations annotated with the stiffness parameter used during teleoperation, and train an imitation policy (state of the art diffusion transformer with image feedback [4]) whose output is an action distribution conditioned on this stiffness label. The key idea is to treat stiffness as an input to the controller, such that compliance becomes a tunable parameter after training and in real time. Evaluation compares manipulation tasks such as pick and place of objects under different stiffness values, in terms of success rate and resulting contact interaction. The project will produce a framework that allows to control compliance in robotic manipulation, which can be integrated into future works where precise force regulation is required.
Milestones
- M1 (Weeks 1-3): Hardware setup; Create the teleoperation framework to collect data.
- M2 (Weeks 4-8): Collect state-action trajectories with stiffness labels on the real hardware, and integrate vision feedback.
- M3 (Weeks 9-12): Train the imitation policy to output positions and regulate compliance.
- M4 (Weeks 13-16): Evaluate for different tasks and configurations
Requirements
- Highly motivated for the topic
- Programming experience (Python)
- Knowledge of Machine Learning / Reinforcement Learning
- Strong mathematical background
- Knowledge of robotics and control theory
Contact
Send an e-mail to both Gregorio Valenti ([email protected]) and Lorenzo Vignoli ([email protected]). To apply, please include a short motivation for this specific project, your CV, and your BSc/MSc transcripts.
Preferences
Between September and December 2026
References
[1] Zeng, Ganwen, and Ahmad Hemami. “An overview of robot force control.” Robotica 15.5 (1997): 473-482.
[2] Hou, Yifan, et al. “Adaptive compliance policy: Learning approximate compliance for diffusion guided control.” 2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025.
[3] Billard, Aude, and Danica Kragic. “Trends and challenges in robot manipulation.” Science 364.6446 (2019): eaat8414.
[4] Chi, Cheng, et al. “Diffusion policy: Visuomotor policy learning via action diffusion.” The International Journal of Robotics Research 44.10-11 (2025): 1684-1704.
[5] https://wuji.tech/en/hand