Lightweight Learning from Actuation-Space Demonstrations via Flow Matching for Whole-Body Soft Robotic Grasping

TL;DR

The paper tackles robust whole-body grasping under uncertainty by leveraging soft robot embodied intelligence and a lightweight actuation-space learning approach. It introduces Rectified Flow to model distributional actuation strategies directly from a small set of demonstrations, avoiding dense sensing and heavy feedback loops. Demonstrated on a tendon-driven SpiRob, the method achieves up to $97.5\%$ grasp success across the workspace from only $30$ demos, generalizes to object size variations of about $\pm33\%$, and remains robust under temporal scaling from $20\%$ to $200\%$ of the reference duration, illustrating strong sim-to-real transfer. Collectively, the work shows that actuation-space learning can convert passive body compliance into functional control intelligence, reducing reliance on centralized controllers for uncertain, contact-rich tasks.

Abstract

Robotic grasping under uncertainty remains a fundamental challenge due to its uncertain and contact-rich nature. Traditional rigid robotic hands, with limited degrees of freedom and compliance, rely on complex model-based and heavy feedback controllers to manage such interactions. Soft robots, by contrast, exhibit embodied mechanical intelligence: their underactuated structures and passive flexibility of their whole body, naturally accommodate uncertain contacts and enable adaptive behaviors. To harness this capability, we propose a lightweight actuation-space learning framework that infers distributional control representations for whole-body soft robotic grasping, directly from deterministic demonstrations using a flow matching model (Rectified Flow),without requiring dense sensing or heavy control loops. Using only 30 demonstrations (less than 8% of the reachable workspace), the learned policy achieves a 97.5% grasp success rate across the whole workspace, generalizes to grasped-object size variations of +-33%, and maintains stable performance when the robot's dynamic response is directly adjusted by scaling the execution time from 20% to 200%. These results demonstrate that actuation-space learning, by leveraging its passive redundant DOFs and flexibility, converts the body's mechanics into functional control intelligence and substantially reduces the burden on central controllers for this uncertain-rich task.

Figures (10)

• Figure 1: Learning from Actuation-Space Demonstration for Grasping. A. SpiRob. B. Distinction in LfD schemes between rigid and soft robots.
• Figure 2: Illustration of the proposed framework. A. Overview of the model architecture designed to learn the flow. B. Training scheme used for optimization. C. Inference scheme applied at test time.
• Figure 3: Example of an expert grasping demo in the simulation.
• Figure 4: Workspace and training region configuration for data generation.
• Figure 5: Experimental setup of the SpiRob platform.