Hasan Poonawala together with Aykut Satici (Boise State University) received a grant from NSF to work on data-driven robotic control using Bayesian Inference and Passivity-Based Control. The project aims to advance the state-of-the-art in contact-rich manipulation by combining planning, control, and machine learning techniques.

Planning algorithms can solve contact-rich manipulation problems when using quasi-static (zero velocity) or quasi-dynamic (constant velocity) models. Executing these plans using standard control techniques has generally been unsuccessful except in simple cases. A key challenge is that actual mode sequences can diverge from the plan very quickly. Techniques to control contact mode transitions are necessary.

The project will combine new methods for data-driven control with computational tools for stability analysis to design controls that stabilize the motion plans by stabilizing contact conditions as needed.