About Us
Manifold Dynamics Research (MDR) focuses on high-fidelity, high-performance fluid-rigid interaction simulations and their applications in the research and engineering of fluid-interactive robotic systems. Established by integrating members from the FLARE Lab and MAgIC Lab at ShanghaiTech University, MDR covers a wide range of research topics including mesoscopic fluid modeling and simulation, high-performance computing, reduced-order modeling, control, decision-making and learning, as well as hardware design, implementation and deployment. MDR aims to advance the research and practical application of sim2real robotic fabrication paradigm, toward the long-term objective of realizing “software defines robotics” in the future.
Gallery
A selection of our work on high-performance simulators and robotics-related applications.
Research
High-performance high-fidelity fluid-rigid interaction simulations
We conduct research and system engineering on high-fidelity mesoscopic fluid simulation models and methods based on the state-of-the-art kinetic approaches. Our work encompasses collision modeling, multi-resolution simulation, turbulence and near-wall modeling, the unlimited-range moving local-domain method, rigid-body system coupling, high-performance system design and implementation, as well as flow-field rendering and visualization. These capabilities enable our simulation system to analyze complex unsteady dynamic fluid-rigid coupling scenarios that traditional simulation software cannot adequately address.
Reduced-order modeling and hybrid simulations
We conduct research on various reduced-order modeling techniques for dynamically two-way coupled fluid-rigid systems, ranging from traditional algebraic models to data-driven approaches. These models are applied to handle the near-field dynamics of rigid bodies, which are integrated with far-field high-performance pure-fluid simulators to form our hybrid simulation system. This enables real-time or near-real-time simulations of complex fluid-rigid interactions on GPU, providing more general and efficient support for the derivation of control policies and decision-making strategies for fluid-interactive robotic systems. The system also supports hardware-in-the-loop validation.
Control, decision-making and reinforcement learning
We conduct research on deriving optimal control algorithms and decision-making strategies using our simulation platforms, exploring both traditional and data-driven techniques. Leveraging advanced simulation capabilities, we aim to automatically synthesize control policies and decision-making strategies that can simultaneously handle both normal and extreme operating conditions for various fluid-interactive robotic systems. These control and decision-making results are expected to be directly transferred to real-world systems, enabling reliable operation with significantly improved success rates.
Hardware design, implementation and deployment
We design hardware systems based on our simulation platform, supporting both geometric shape and control optimization. Our hardware prototypes are fabricated via 3D printing, with control algorithms physically implemented on these systems. We have successfully developed various fluid-interactive hardware platforms, ranging from robotic fish to drones, to validate our “software defines robotics” concept.
Platform
We provide a suite of platforms for fluid-rigid interaction simulation and robotics. A selection of these platforms is presented below.
Contact
For collaborations, inquiries, or any other opportunities, please contact us via email.
Email: contact@manifolddynamics.com