Contact Manipulation
estimation, planning, and control through contact
Contact manipulation is an rising topic in robotics that involves solving various inverse problems related to contact. Our focus is on developing efficient and scalable methods for these problems, leveraging model-based optimization, sampling techniques, and integrating data-driven approaches.
Contact Factor Graph
Graphical modeling is widely used to represent problems in a structured manner with a stochastic perspective. The Contact Factor Graph (CFG) framework enables versatile reasoning about contact interactions between objects through differentiable, compositional factor modeling and an efficient gradient-based inference algorithm.
Here is a simple illustrative example:
Shaping Solution Landscape
An often overlooked perspective in model-based approaches is the impact of a well-designed model on the entire solution process. A good model is not solely about accuracy; it also shapes a problem landscape that is easier to solve, guiding towards reasonable solutions more efficiently. We are exploring techniques to adapt models into “easier-to-solve” forms to address challenging problems. A notable example is homotopy-based narrow passage robot planning, which incrementally transforms the environment model.
