First-Principles Models and Safety-Driven Planning for Soft and Rigid Robots

Abstract

Humans naturally blend softness and rigidity in both their physical structure and control strategies, enabling them to physically interact with the world in ways that balance capability and safety. Replicating this balance in robotic systems is challenging, as most designs are predominantly soft or rigid, each with trade-offs. As robots integrate into human-centric environments, ensuring safe and effective physical interaction is paramount. In robotics, physical interaction encompasses functional tasks like manipulation and communicative roles like haptic feedback. Failures in these interactions pose risks ranging from miscommunication to equipment damage or even human injury. Different robot morphologies present different challenges for ensuring safe and effective interaction. Rigid robots excel at precise manipulation, but their reliance on accurate modeling complicates safety guarantees when faced with uncertainties and real-time constraints. In contrast, soft robots naturally offer compliance that enhances safety but suffer from limited force transmission and complex modeling challenges. This thesis focuses on developing tractable models and optimization frameworks that aim to help robotic systems navigate this safety-functionality trade-off more effectively.

Rigid robotic manipulation faces significant challenges in dynamic, real-world environments, where safety, autonomy, and robust handling of model uncertainty are critical. Current approaches experience a trade-off between real-time operation and safety during operation, making them unsuitable for human-centric environments. To address this trade-off, I developed WAITR (Wrench Analysis for Inertial Transport using Reachability), a real-time and provably safe planning and control framework for non-prehensile manipulation of unsecured objects under model uncertainty. Like a restaurant waiter balancing a tray, WAITR determines how to safely apply contact wrenches to manipulate unsecured objects. Reachability analysis is used to conservatively overapproximate contact wrenches enabling both continuous-time safety and robust accounting for model uncertainty. WAITR is validated through simulation and hardware experiments.

Physical interaction is an important form of communication that can enhance virtual and augmented reality experiences and improve access to information for blind and visually impaired individuals. One method of communication is through haptic shape displays, which communicate tactile information through a deformable surface. Current displays struggle with scalability, limited resolution, large form factors, and an inability to independently render shape and stiffness. A promising method of actuation is pneumatics, which is highly scalable and can be used in soft structures to render surfaces, though force transmission for rendering stiff objects is lacking. To address current limitations, the use of articulated inflatable mobile structures (AIMS) as an actuation technology was investigated for improving haptic shape displays. Through encasing soft inflatable actuators with rigid linkages, AIMS can maintain a high mobility while improving the force transmission of the soft inflatables. A network-based model was derived and experimentally validated for single-cell AIMS. Then, the modeling framework was extended to multi-cell AIMS where shared structural elements enable coupled cells to affect each other through transmitting forces and motions. It was demonstrated that multi-cell AIMS can independently control both their shape and stiffness through appropriate selection of the control inputs. Finally, two optimization problems were formulated. The first determines the pressure inputs required to render a desired shape. The second co-optimizes the structural design and control inputs to achieve specific shapes. These optimization frameworks demonstrate the viability of using AIMS for haptic shape displays.

This dissertation introduces tractable frameworks for safe manipulation and expressive haptic feedback, enabling safer and more capable systems for both manipulation and haptic communication.