Outline
This project aims to develop an accurate controller for soft robotic arms [1] that achieves reliable trajectory tracking across varying spatial poses and payload conditions. The project will explore control methods, including (1) model predictive control (MPC) based on identified or neural network models; (2) reinforcement learning methods with domain randomization or adaptation.
Successful project completion may get a chance to attend the 2026 ICRA WBCD competion [2].
Milestones
- Weeks 1–2: Soft robot arm setup, sensing, and control interface integration.
- Weeks 3–4: Data collection across multiple poses and payload conditions; baseline model identification.
- Weeks 5–10: Implementation of MPC and RL-based controllers with domain randomization/adaptation.
- Weeks 11–14: Experimental evaluation of tracking accuracy, robustness to pose changes and various payloads.
Requirements
Proficient in Python and independent troubleshooting; Experience with ROS and real-robot experiment; Knowledge/experience with the implementation of MPC and RL methods.
To Apply
Please email (1) a short paragraph describing your background, and (2) your transcripts to:
[email protected] and [email protected].
Reference
[1] Soft robot arm: https://ieeexplore.ieee.org/abstract/document/11049035
[2] WBCD Competition Website: https://wbcdcompetition.github.io/