Modular reconfigurable robots capable of programmable shapes and motions

Abstract

For robots to successfully locomote in different environments, it's better to equip them with multiple modes of locomotion.Multimodal locomotion for existing robotic systems is generally realized by integrating multiple mechanisms into a single robot. These reconfigurable robots are generally cumbersome or challenging to control and actuate. The adaptability of these robots is also limited after fabrication. Also, in the field of soft robotics, shape morphing structures actuated by external stimuli are utilized to enable reconfigurable and multi-functional robots. Since the soft robots feature continuous and large deformation, the behavior of these robots is not accurately predictable or controllable. We propose to realize the reconfigurable robots by stacking multiple reconfigurable modules in series and using a few tendons for actuation. The shape and motion of the reconfigurable modules can be adjusted upon tendon actuation. These modular reconfigurable robots offer advantages such as ease of fabrication, extensive workspace, predictable behavior, and simplified control. First, we present a novel reconfigurable module inspired by the Kresling origami pattern. The origami module is equipped with joints that can independently transition between soft and rigid states, enabling the module to adapt its behavior during actuation. To understand the reconfiguration capability, we numerically analyze the programmable shapes and motions of a single origami module. We develop a reconfigurable robot with four legs, each made from four serially connected modules. The robot can walk, crawl, and inch using the same mechanical structure. We also realize a reconfigurable robot based on reconfigurable bistable module. Bistable modules have the ability to rapidly switch between multiple stable states and hold any stable state without requiring external energy input. Traditional bistable modules generally have a fixed structure after fabrication, leading to a fixed behavior when actuated. To make the bistable modules reconfigurable, we investigate adjusting the behavior of the bistable modules with shape morphing beams and adjustable springs. A forward model to predict the module's behavior is presented and validated. We also address the inverse problem to achieve a desired behavior by choosing proper parameters of the reconfiguring process. By stacking two modules in series, we realize a reconfigurable arm capable of generating different trajectories and a reconfigurable crawler which can crawl in different directions. Both the reconfiguring parameters (e.g., stiffnesses of the joints of the origami modules) and the control algorithm (tendon displacements) will influence the resulting shape and motion of the robots. We manipulate the origami-module based and bistable-module based reconfigurable robots by empirically choosing the appropriate reconfiguring parameters and control algorithm. For more complicated task, the reconfiguring parameters and control algorithm should be co-optimized, since the reconfiguring parameters would influence the kinematics or dynamics of the robots, resulting in varying control algorithms. Here, we employ reinforcement learning based co-optimization framework on the robot to obtain an optimal combination of reconfiguring parameters and corresponding control policy. The task investigated in this work is to make the end effector of a manipulator composed of multiple serially connected origami modules reach a specific goal point, with optional obstacle avoidance. We further deploy the learned parameters and control policy on a two-module prototype performing a reaching task to validate the effectiveness of the proposed co-optimization framework. This work demonstrates a unified approach to designing and controlling modular reconfigurable robots, paving the way for future robotic systems capable of autonomous adaptation and multifunctional operation.