CAREER: Soft Robotic Fingertips with High-Resolution, Calibrated Shape and Force Sensing for Dexterous Manipulation

Stanford University, Stanford CA

Investigators

Abstract

Robots capable of manipulating objects like humans will greatly benefit society from elderly care in assisted living to advanced manufacturing, such as small component assembly. The current key barrier to robots performing these types of service tasks is insufficient dexterity. To approach human dexterity, robots must be able to move their fingers and sense their touched surroundings with a comparable level of resolution to humans. The key challenge is providing high-resolution, calibrated tactile sensing, which once modeled can be leveraged for advanced robotic manipulation. This Faculty Early Career Development (CAREER) grant supports a sensor design that leverages a camera to observe the contact boundary of a transparent robotic fingertip. When the camera observes deformation of the soft robotic fingertip from touching an object, these observed deformations are mapped to the shape of the deformed finger and applied forces on the fingertip surface through the novel modeling technique. The calibration and modeling techniques to be developed in this project allow for manipulation motion planners to determine the best sequences of grasps to reposition an object in the robot’s hand with an estimate of expected manipulation success. This will advance the field of robotics by increasing dexterous manipulation ability in service tasks benefiting the US economy, prosperity and welfare. The research in robotic perception and manipulation supported by this grant will promote education in robotics. Through close involvement with non-profit organization “Black In Robotics” the research team will also broaden the participation of underrepresented groups as students learn to design and research robotic systems capable of performing advanced service tasks. The optical tactile robotic fingertip is modeled as a Cauchy elastic material with a direct correlation between the observed strain and stress field of the sensor. By illuminating the interior of the sensor with multi-colored light-emitting diodes (LEDs), the contact surface shape is reconstructed by correlating light intensity to the interior surface normal, and strain is observed through the deformation of contact surface markings. The novel calibration method enables high-resolution measurement of both the robotic finger contact surface shape and stress field. This is then used to model grasp stability through the high-resolution limit surface and position within the limit surface for a multiple soft-finger grasp. This representation is then leveraged by a high-level grasp planner to strategize a sequence of grasps for in-hand manipulation between target object poses. The modularization of the manipulation problem is expected to increase the rate of motion planning adaption to novel objects while providing a more accurate estimation of grasp stability. This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.