Surface-Based Manipulation with Modular Foldable Robots

TL;DR

The paper tackles the challenge of manipulating fragile, deformable, and irregular objects without traditional grasps by introducing surface-based manipulation using modular foldable robots that form reconfigurable planar end-effectors. It presents three manipulation primitives—translation, rotation, and flipping—demonstrated on soft, rigid, and deformable objects with vision-based closed-loop control and a hierarchical planning architecture. The work provides a scalable modular design, detailed perception and control pipelines, and quantitative validations, highlighting potential applications in packaging and delicate material handling. Overall, this surface-centric approach expands robotic manipulation capabilities beyond grasp-based methods and offers a pathway toward robust, geometry-aware automation across diverse object classes.

Abstract

Intelligence lies not only in the brain (decision-making processes) but in the body (physical morphology). The morphology of robots can significantly influence how they interact with the physical world, crucial for manipulating objects in real-life scenarios. Conventional robotic manipulation strategies mainly rely on finger-shaped end effectors. However, achieving stable grasps on fragile, deformable, irregularly shaped, or slippery objects is challenging due to difficulty in establishing stable forces or geometric constraints. Here, we present surface-based manipulation strategies that diverge from classical grasping approaches, using flat surfaces as minimalist end-effectors. By adjusting surfaces' position and orientation, objects can be translated, rotated, and flipped across the surface using closed-loop control strategies. Since this method does not rely on stable grasping, it can adapt to objects of various shapes, sizes, and stiffness levels and can even manipulate the shape of deformable objects. Our results provide a new perspective for solving complex manipulation problems.

Figures (6)

• Figure 1: Surface-based manipulation. (a) Object properties influence whether we intuitively grasp or support them with a surface. Small, rigid objects with defined edges are easily grasped, while large, round, deformable, soft, or slippery objects are more effectively manipulated through surface contact. This distinction highlights how surface-based manipulation complements traditional grasping in handling diverse objects. Building upon this, we present a novel approach to robotic object manipulation using surfaces. This is achieved through the integration of three main motion principles: (b) translation, (c) flipping, and (d) rotation. (e) An example is in food packaging automation, where it addresses the challenge of handling items with varied shapes without causing damage. Surface-based manipulation enables tasks like rotating or positioning grape bunches for packaging and inspection.
• Figure 2: Control schematic for surface-based object manipulation. (a) High-level controller involves five main components. Main manipulation configuration includes two modules side by side. The vision system captures the object’s position and orientation and provides real-time feedback to task planner. (b) The task planner, a state machine, defines surface orientations and height and then these are converted into joint reference positions via inverse kinematics. It transitions among three tasks: (i) translation, (ii) flipping, (iii) and rotation. Translation adjusts the surface's roll and pitch to move the object, followed by height oscillations. Rotation has two phases: Phase 1 adjusts Surface 2's roll, and Phase 2 raises and lowers it to repeat Phase 1 until the desired angle is achieved. Flipping occurs through the rapid elevation of Surface 2. (c) The low-level controller uses BLDC motors with a servo position controller to track joint angle commands with real-time feedback.
• Figure 3: Surface-based manipulation of various types of objects. A rigid object (a solid, uniform square prism block) and the first two white soft objects (a malleable slime) are manipulated using closed-loop control, while the remaining objects are manipulated via open-loop teleoperation to transition between different states, demonstrating the strategy’s versatility across diverse object types. (a) Rigid object transfer (T), rotation (R), and flipping (F). (b) Manipulation of soft and deformable objects: adjusting the position or shape of deformable materials in various forms. (c) Manipulation of different objects, from top left to bottom right: a cookie, a cake in a transparent package, a cotton-stuffed toy fish, a roll of tape, and a bag of unevenly distributed popcorn.
• Figure 4: Translation of soft objects. In this experiment, the efficacy of our translation strategy was assessed using a deformable object positioned at multiple initial locations. (a) Initial setup for three sequential tests with object: Lighter colors indicate initial positions and darker colors show target positions. Objects are manually reset to new initial positions after each translation. (b) The object's trajectory in the XY-plane and surface adjustments in roll, pitch, and height are controlled by the transfer strategy.
• Figure 5: Validation of translation, flipping, and rotation strategies for rigid objects. (a) In translation and flipping experiments, the object (measuring $8 \times 8 \times 3\ cm^3$ and weighing $110g$) translates from an initial point on Surface 1 to between two surfaces and then flips. After flipping, the object's position shifts towards Surface 1, and it is translated again between two surfaces. (b), the position of the object in the surface frame is depicted over time. A noticeable $180\degree$ transition in the object’s yaw orientation in the global frame is attributed to the flipping motion. Additionally, changes in the height of Surface 1 and Surface 2 during manipulation are shown.(c) The schematic shows the rotation experiment setup with the object positioned between the two surfaces. The experiment sets four reference angles of $20\degree$, $30\degree$, $60\degree$, and $80\degree$ and aims to rotate the object to the target orientation. (d) The reference angle and the object's actual orientation are compared. It also highlights the dynamic adjustments in Surface 2's roll and height during operation.