GenORM: Generalizable One-shot Rope Manipulation with Parameter-Aware Policy

TL;DR

GenORM tackles the data inefficiency of deformable rope manipulation by introducing a parameter-conditioned policy that generalizes across rope dynamics with a single real-world demonstration. It trains the policy in simulation with diverse rope parameters (Young's modulus and Poisson's ratio) and employs a gradient-based Real2Sim estimator that minimizes a grid-density loss on a differentiable physics simulator to recover those parameters at test time. The approach yields strong in-domain and out-of-domain generalization in simulation and real robots, outperforming baselines by substantial margins and reducing real-world data requirements. This work advances practical one-shot manipulation of deformable objects and highlights a viable path for parameter-aware policies and gradient-based system identification in robotics.

Abstract

Due to the inherent uncertainty in their deformability during motion, previous methods in rope manipulation often require hundreds of real-world demonstrations to train a manipulation policy for each rope, even for simple tasks such as rope goal reaching, which hinder their applications in our ever-changing world. To address this issue, we introduce GenORM, a framework that allows the manipulation policy to handle different deformable ropes with a single real-world demonstration. To achieve this, we augment the policy by conditioning it on deformable rope parameters and training it with a diverse range of simulated deformable ropes so that the policy can adjust actions based on different rope parameters. At the time of inference, given a new rope, GenORM estimates the deformable rope parameters by minimizing the disparity between the grid density of point clouds of real-world demonstrations and simulations. With the help of a differentiable physics simulator, we require only a single real-world demonstration. Empirical validations on both simulated and real-world rope manipulation setups clearly show that our method can manipulate different ropes with a single demonstration and significantly outperforms the baseline in both environments (62% improvement in in-domain ropes, and 15% improvement in out-of-distribution ropes in simulation, 26% improvement in real-world), demonstrating the effectiveness of our approach in one-shot rope manipulation.

Figures (4)

• Figure 1: Overview of our Pipeline: (a) We randomly sampled the parameters and set it to simulation where we collect release coordinates and goal coordinate. Receiving dataset, a simple MLP policy is trained. (b) For inference, deformable object parameters on simulation are estimated from real-world objects by taking loss from the grid density of point clouds between simulation and real-world via differentiable physics. (c) We transfer the policy onto real robots and conduct tests.
• Figure 2: (a) Three different type ropes; cotton rope (red), polyester rope (white), and polyethylene rope (yellow). The estimated Young’s modulus and Poisson’s ratio are [1779.38, 0.35], [3276.12, 0.346], and [8000.31, 0.36] for red, white, and yellow rope respectively. (b) Our experiments setup for Real2Sim: two Intel Realsense Depth Camera D435i and xArm 7.
• Figure 3: As a qualitative assessment, we visualize rope trajectories in the real world and simulation with estimated parameters (Table \ref{['tbl:real2sim']}) from the three different release points, upper, middle, and lower.
• Figure 4: The real-world deployment evaluation for red rope. For each policy, left figure proposes release coordinates (2-dimensions) and right figure illustrates the test performance for three trials. Each color is unified according to a goal coordination. The colors on the left and right sides of the figure are determined based on the goal coordination.