DynDLO: Learning-Based Trajectory Planning for Dynamic Robotic Manipulation of Deformable Linear Objects
Daniele Maria Liuni, Alessandro Bartesaghi, Andrea Monguzzi, Alessandra Miuccio, Andrea Maria Zanchettin, Paolo Rocco
AI summary
Problem
Dynamic manipulation of Deformable Linear Objects (DLOs) remains under-explored because most prior work assumes quasi-static, low-speed conditions, while model-based methods struggle with complexity and computational cost.
Approach
DynDLO is a goal-conditioned, robot-independent Reinforcement Learning sandbox that uses B-spline trajectory parametrization and tailored reward functions to train agents for implicit and explicit DLO shape control tasks in MuJoCo simulation.
Key results
- Introduces a 6-axis robot-independent RL sandbox for dynamic DLO manipulation.
- Develops a B-spline-based trajectory generation strategy enabling complex single- and multi-step policies with a low-dimensional action space.
- Designs task-specific reward functions for implicit and explicit shape control tasks.
- Demonstrates successful sim-to-real transfer across four dynamic tasks without real-world training.
Why it matters
It provides a flexible, computationally efficient framework for training robots to handle fast, dynamic cable manipulation, bridging a critical gap in soft-object robotics and reducing reliance on costly real-world learning.
Abstract
The automatic manipulation of Deformable Linear Objects (DLOs) remains currently a challenge in robotics. Previous research on robotic DLOs manipulation has primarily addressed quasi-static DLO manipulation at low speeds, leaving the potential of dynamic DLO manipulation largely unexplored. This paper introduces DynDLO, a goal- conditioned, 6-axis robot-independent Reinforcement Learning sandbox for training agents on a variety of DLO dynamic manipulation tasks. In DynDLO, a DLO attached to the robot Tool Center Point (TCP) is simulated in the MuJoCo environment. By employing a B-Spline based trajectory generation function, the agent is capable of learning single and multiple step trajectories for the TCP, which succeed in various DLO dynamic manipulation problems. Specifically, we propose tailored design strategies for the reward function according to the classification of tasks into implicit or explicit DLO shape control tasks. Experiments on four representative tasks demonstrate that DynDLO is capable of generating dynamic manipulation policies that transfer successfully from simulation to the real world, achieving high success rates without requiring real-world training. Link to the video: https://youtu.be/-W3tKXyenO4 Link to GitHub page1