Search papers, labs, and topics across Lattice.
This paper introduces a novel approach to bridging the sim-to-real gap in robotics by utilizing renormalization to define effective, resolution-dependent parameters that account for omitted details in simulators. By analytically demonstrating this mechanism in proportional-derivative (PD) control, the authors reveal how feedback alters effective gains and inertia, providing insights into dynamic tasks like rope manipulation and underwater swimming. The proposed method not only enhances the understanding of real behavior in robotic systems but also offers a practical framework for selecting observables and identifying missing physics in simulations.
Renormalization can redefine how we understand and address the sim-to-real gap in robotics by leveraging effective parameters that capture omitted dynamics.
Bridging the sim-to-real gap is a central problem in robotics, and the prevailing approach is to build increasingly accurate simulators. Here, we propose another approach based on renormalization: using effective, resolution-dependent parameters to absorb details omitted by the simulator and reproduce real behavior. These parameters may differ from measured physical values because they compensate for what the simulator leaves out. We demonstrate this mechanism analytically for proportional--derivative (PD) control at finite simulation frequency, where proportional feedback changes the effective derivative gain and derivative feedback changes the effective inertia. We then interpret dynamic rope manipulation and underwater swimming through the same perspective. Finally, we present a practical procedure for choosing observables, identifying omitted physics, and determining effective parameters. Renormalization offers robotics a complementary path across the sim-to-real gap: effective parameters, real behavior.
Stanford HAI