Hardware, control and perception for robust manipulation in cluttered environments
Toyota's Woven city aims to build residences that include novel technologies such as robots to assist with daily living. This project contributes to this vision by enabling a robot to perform pick and place tasks in complex, cluttered environments. Specifically, we consider common scenarios in the home which are characterised by high amounts of clutter and thereby instantiate fundamentally hard manipulation problems.
In this project, we are addressing challenges such as restricted workspaces and a high intra and inter-class variety of objects by integrating advanced tactile sensing and novel robot hands with grasp and manipulation policies that fuse vision and touch sensing. This new hardware provides an opportunity to make contact-rich manipulation robust, by characterizing contact conditions and providing dexterity in a confined workspace. Additionally, new sensors require new approaches to wiring and integration, which we plan to address by utilizing new bonding techniques to connect stretchable sensors to microprocessors for filtering, communication and local control.
Another fundamental goal of this project is developing manipulation strategies that exploit environmental constraints by making deliberate contact, by funneling uncertainty and thereby making manipulation more robust. To this end, we are investigating novel approaches with learned high-level pick and place planning that orchestrates low-level reflexes.