Research

Simulation model and framework (ICRA 2022)

Model-based full-shape controller (RoboSoft 2022)

Integrated power converter and control circuits (APEC 2022)

Soft robots have garnered significant interest because of their ability to take on rich shapes and motions. Piezoelectric soft robots, in particular also bring the benefits of fast actuation, large forces, and small form factors. However, the strong sensitivity to external forces makes the statics and dynamics of soft-robot motions, challenging to model. This work explores the integration of sensing, powering, and control circuitry on the robot, which introduces complex force interactions and imbalances.

Our work develops a pseudo-rigid body model for scalable simulations, a model-based full-shape controller, and integrates power and control electronics onto a robot platform for battery-powered, wirelessly-controlled, untethered operations. A five-actuator planar soft robot is constructed with piezoelectric devices bonded to a layer of steel foil. The simulation framework employs discrete rigid-link elements connected by motors to model the actuators with a physics engine PyBullet. We validate the simulations for both static and dynamic analysis of a single-actuator cantilever and also for the complete five-actuator robot. This includes static shapes, quasi-static inchworm motion, and dynamic jumping. The robot can jump as high as 20x of its height and move as fast as 6 cm/s, by optimizing its weight distribution. We further develop and demonstrate a model-based controller for the robot to crawl under overhead constraints. The controller uses a soft-body continuous model for shape planning and control. Finally, we present our recent work on untethered piezoelectric soft robots, including the design and integration of customized power electronics and control circuits for battery powering and wireless control, with the robot exhibiting maneuverable bidirectional dynamic crawling. Future work includes exploration of model-free reinforcement learning for robot motion control.

Conference Paper at ISSCC 2019

Invited Journal Paper at IEEE TBioCAS

Our work develops an architecture for high-spatial-resolution tactile-sensing skins based on LAE-CMOS hybrid systems, where compressed sensing (CS) is used to reduce interfacing complexity according to signal sparsity. A resistive force-sensing array with 20 elements is demonstrated, using ZnO TFTs to provide CS acquisition and a custom CMOS IC to provide readout/control. Acquisition error of 0.7kΩRMS is achieved over a 100kΩ-15kΩ sensing range, at energy and rate of 1.2μJ/frame and 31fps.