Sling2Sim2Real: One-Shot Elastic System Identification for Non-Destructive Slingshot Policy Learning

2026-07-25

AI Summary

Manipulating deformable objects in the real world is challenging due to their high-dimensional nonlinear deformations and material properties that are difficult to observe visually, leading to high costs and frequent damage during policy learning. This work proposes a framework requiring only a single non-destructive real-world interaction: it efficiently identifies elastic parameters through multi-start Real2Sim system identification combined with parameter covariance analysis, then leverages simulation-based reinforcement learning and zero-shot Sim2Real policy transfer to deploy policies without additional real-world data. Evaluated on a slingshot task using a Franka robotic arm, the method drastically reduces the number of real-world trials while achieving high-precision control and strong generalization.

Abstract

Elastic object manipulation (EOM) involves highdimensional, nonlinear, and elastic deformations. The diverse deformation properties of elastic objects substantially expand the relevant state space, requiring extensive exploration to learn accurate manipulation policies for tasks such as slingshot manipulation. While simulation enables large-scale and safe exploration compared to costly and potentially destructive real-world trials (e.g., repeated projectile launches), accurately calibrating elastic behavior between the real world and simulation remains challenging since elastic properties are largely indistinguishable from visual observations alone. To address these challenges, we propose Sling2Sim2Real, a one-shot Real2Sim2Real framework that identifies elastic parameters from a single non-destructive interaction and enables policy learning in simulation. The framework consists of two stages: 1) a multi-start Real2Sim system identification method that exploits parameter covariance to estimate elastic properties, and 2) simulation-based policy learning followed by zero-shot Sim2Real transfer using the calibrated simulator. We evaluate Sling2Sim2Real on a slingshot manipulation task using a Franka Emika Panda arm and elastic bands with diverse physical properties across varying target distances. Experimental results demonstrate that Sling2Sim2Real achieves accurate policy learning and robust generalization while significantly reducing the amount of required real-world interaction.

Problem

• elastic object manipulation
• system identification
• sim2real transfer
• slingshot policy learning
• non-destructive interaction

Innovation

• one-shot system identification
• elastic object manipulation
• Sim2Real transfer
• non-destructive interaction
• parameter covariance