Combining and Decoupling Rigid and Soft Grippers to Enhance Robotic Manipulation

Abstract

For robot arms to perform everyday tasks in unstructured environments, these robots must be able to manipulate a diverse range of objects. Today’s robots often grasp objects with either soft grippers or rigid end-effectors. However, purely rigid or purely soft grippers have fundamental limitations as follows: soft grippers struggle with irregular heavy objects, whereas rigid grippers often cannot grasp small numerous items. In this article, we therefore introduce RISOs, a mechanics and controls approach for unifying traditional RIgid end-effectors with a novel class of SOft adhesives. When grasping an object, RISOs can use either the rigid end-effector (pinching the item between nondeformable fingers) and/or the soft materials (attaching and releasing items with switchable adhesives). This enhances manipulation capabilities by combining and decoupling rigid and soft mechanisms. With RISOs, robots can perform grasps along a spectrum from fully rigid, to fully soft, to rigid-soft, enabling real-time object manipulation across a 1.5 million times range in weight (from 2 mg to 2.9 kg). To develop RISOs, we first model and characterize the soft switchable adhesives. We then mount sheets of these soft adhesives on the surfaces of rigid end-effectors and develop control strategies that make it easier for robot arms and human operators to utilize RISOs. The resulting RISO grippers were able to pick up, carry, and release a larger set of objects than existing grippers, and participants also preferred using RISO. Overall, our experimental and user study results suggest that RISOs provide an exceptional gripper range in both capacity and object diversity.

Introduction

Across food processing, parts manufacturing, and assistive caregiving, there is a need for robot arms that can grasp and manipulate a diverse range of everyday objects.^1–4 Industry-standard arms often use rigid grippers that pinch items between two or more nondeformable fingers.^5–9 However, these rigid grippers are not able to pick up and hold objects that are small, numerous, or irregularly shaped (e.g., a pile of candy). To address these shortcomings, recent works have developed a variety of soft grippers using mechanisms such as gecko-inspired adhesives,^10–14 granular jamming,^15,16 electroadhesion,^17–19 or responsive materials.^20–26 But soft grippers have a separate grasping domain and are often unable to hold the large heavy items that rigid end-effectors are able to pick up and carry (e.g., a bottle of syrup). When a robot arm leverages grippers that are either purely rigid or purely soft, it fundamentally limits the types of objects that robot can manipulate.

We therefore seek to expand the range of graspable objects by combining rigid and compliant components within the robot’s gripper. Recent works have started to develop grippers that integrate both elements.^27–35 However, within existing designs the rigid and soft capabilities are inherently coupled; for example, robotic fingers that alternate between rigid links and soft joints. By contrast, our core insight is that we can decouple and couple the rigid and soft mechanisms by combining traditional rigid end-effectors with soft adhesive sheets. This decoupling has the potential to fundamentally increase the range of robotic grasps as follows: from purely rigid (e.g., manipulating a bottle of syrup) to purely soft (e.g., picking up a pile of candy) to a combination of rigid and soft (e.g., holding beads of multiple sizes).

We leverage our insight to create RISOs, robotic grippers that unify RIgid end-effectors with SOft materials (see Fig. 1A, B). RISOs are formed by mounting sheets of switchable adhesives onto the surfaces of rigid robotic fingers and then controlling both the rigid and soft components during manipulation tasks. When the robot arm moves to pick up an item, it can (1) pinch the object between the fingers of the rigid gripper, (2) cause the object to stick to the soft surfaces, or (3) use a combination of the rigid pinch and soft adhesion to hold the object (see Fig. 1C). In practice, RISO advances the range of robotic grasping by combining and decoupling purely rigid and purely soft grippers. For example, RISOs can grasp objects in real time across a 1.5 million times range in mass—picking up, carrying, and releasing objects from 2 mg to 2.9 kg (see Fig. 1D, Supplementary Figs. S6 and S7).

FIG. 1.

RISO enhances grasping by combining and decoupling rigid and soft mechanisms. (A) Human operators and robot arms can leverage RISOs to pick up, hold, and release objects. (B) RISOs are formed by mounting soft adhesive sheets to the surfaces of traditional rigid end-effectors. (C) When grasping an item RISO can use a fully rigid grasp (pinching the object between nondeformable fingers), a fully soft grasp (causing the object to adhere to its surface), or a combined rigid and soft grasp. (D) With this spectrum of grasps, RISO is able to pick up objects ranging from 2 mg items to 2.9 kg, a 1.5 million times change in mass.

We first explored the feasibility of RISO in preliminary work.³⁶ In this article, we now make the following contributions toward a unified framework for RISO grippers:

Characterizing gripper capabilities

Functionally, our soft materials take the form of flat silicone sheets that we can rapidly control to switch between grasping and releasing in less than 0.1 s. We characterize the gripping force applied by these structures as a function of the grasped object’s radius, curvature, roughness, and porosity. We find that the soft materials combine both adhesion (i.e., using surface forces to attach to the object) and wrapping (i.e., constricting around the object), achieving forces reaching 50 N in idealized testing conditions.

Integrating RISOs with robot arms and human users

We recognize that increased gripping capability is not meaningful if robot arms and human operators cannot harness RISOs. Accordingly, we develop control strategies with varying levels of autonomy to make it easier for robots and humans to utilize RISOs. In particular, we present a shared autonomy approach that (1) uses the human’s joystick inputs to infer their desired object and grasp type and then (b) autonomously aligns the RISO to complete the intended grasp. We show that this shared autonomy method reduces the number of human inputs needed to complete manipulation tasks.

Comparing to existing grippers

We perform grasping experiments with robot arms and human operators to compare RISOs, an industrial soft gripper, and a granular jamming gripper. Across both autonomous and human-controlled conditions, we observe that robots equipped with RISOs are able to pick up, hold, and release a more diverse range of everyday objects. User responses to Likert scale surveys indicate that our participants subjectively rated the RISO gripper as easier to user and more preferable than the tested state-of-the-art alternatives.

Demonstrating practical applications

We finally showcase our system’s practical abilities in assistive scenarios by controlling a robot arm and RISO gripper to make pizza. Here the human teleoperates the arm and RISO to manipulate larger food objects (e.g., placing the crust, spreading the sauce) and arrange smaller food items (e.g., pepperoni, pepper, and olive toppings). When viewed together, our results suggest that RISOs provide an exceptional gripping range in both capacity and object diversity, clearly distinguishing our framework from the baselines. These advances provide a path forward for robotic grippers, particularly within unstructured environments where manipulation of diverse and unexpected items is essential. See videos of our RISOs and experiments at https://youtu.be/du085R0gPFI.

Materials and Methods

Here we present our mechanics and control approach for RISO grippers. In the “Creating RISO grippers” section, we explain the physics behind the soft adhesives, as well as their fabrication process. We then form RISOs by mounting sheets of these pneumatically actuated soft materials onto the surfaces of rigid end-effectors. This combination of a rigid gripper and adhesive sheets introduces new control variables as follows: for example, deciding whether to use the soft adhesives or rigid grippers to pick up an object. In the “Controlling RISO grippers” section, we partially automate these decisions and make it easier for humans and robot to control RISOs by developing a share autonomy approach for RISO grippers.

Creating RISO grippers

To understand the adhesive capacity of the soft gripper, we start with the underlying physical relationship between adhesive force F_c, contact area, and gripper compliance as follows^37,38: $F_{c} \sim \sqrt{G_{c}} \sqrt{\frac{A}{C}}$ (1)

Here A is the contact area between the soft gripper and the target object, C is the gripper’s compliance in the loading direction, and G_c captures the fracture energy of the interface. In practice, G_c can be treated as the energy per unit area that needs to be applied to separate the interface between gripper and object. In the case of reversible adhesives that do not rely on covalent chemical bonds,^39–41 higher G_c can be achieved through viscoelastic dissipation or other lossy processes at the interface.

For object manipulation, it is essential to be able to pick up and then subsequently release diverse objects. One challenge with soft grippers that utilize adhesion is to make an adhesive strong enough to pick up an object but weak enough so that the object can be released. To pick up a wide range of objects, from small delicate items to large heavy objects, adhesives have to achieve strong and releasable adhesion within the same material. So the range of mass that a gripper can manipulate becomes a key metric. Therefore, of particular interest for adhesive-based object manipulation is the ability to dramatically increase F_c when gripping an object and then rapidly decrease F_c when releasing that object. To maximize F_c, Equation (1) demonstrates that A should be increased, C should be decreased, and G_c should be high. Typically, to achieve high contact area A between the gripper surface and the surface of the object, the gripper should be as soft as possible. However, to minimize compliance C, the gripper also needs to be as stiff as possible. To accommodate these contrasting design directions, we focus on the overall A/C ratio between contact area and compliance. We propose an adhesive design that is soft during initial contact between gripper and object (increasing contact area A) and then becomes rigid while maintaining that same contact area (decreasing the compliance C). Accordingly, to achieve a large A/C ratio, it is desirable for the adhesive to be tunable so that the compliance can be changed during contact to first increase A and then decrease C.

We apply these principles to create the soft adhesive for RISO grippers. This adhesive consists of a tunable elastomeric membrane on a foam foundation (see Supplementary Fig. S1 for fabrication details). We actively control the membrane through pneumatic pressure (see Supplementary Fig. S2), allowing for three primary states as follows: neutral pressure where the membrane is flat, positive pressure when the membrane inflates, and negative pressure where the membrane is constricted (decreased compliance). Grasping and releasing items is achieved by switching between these states.

Release

To release an item, we want to minimize F_c. Our soft adhesive achieves this by switching from negative pressure to positive pressure and inflating the membrane (Supplementary Fig. S3). This inflation causes the contact area to decrease, minimizing the ratio A/C.

Neutral to negative

To increase F_c and adhere to an object, our soft gripper can use two different processes. The first is neutral to negative, where the membrane is initially at atmospheric pressure. We start the grasp by bringing the membrane into contact with the object and then apply a negative pneumatic pressure to increase the stiffness of the membrane and foundation. This decreases C while having a negligible effect on contact area A, increasing gripper force F_c (see Fig. 2A).

FIG. 2.

Measuring the force capacity of RISO’s soft adhesives. (A) Soft grasps while switching the membrane from neutral-to-negative pressure. (B) Soft grasps while switching the membrane from positive-to-negative pressure. (C) Force profiles for neutral to negative and (D) positive to negative with a 12.5 mm smooth indenter (circles represent testing stages from A and B) (E) Force capacity F_c vs. indenter radius. (F) Force capacity F_c vs. $\sqrt{A / C}$ , where the points represent the experimental data and the lines represent the prediction from Equation (1). Here G_c = 4.2 J/m $^{2}$ and G_c = 44.7 J/m $^{2}$ for the lower and upper lines. (G) Adhesion switching ratio (SR) as a function of indenter radius, where F_low are data from positive to positive.

Positive to negative

To further increase the grasping force, we can alternatively start the membrane at a positive pressure (i.e., the membrane is initially inflated). Here the membrane is first inflated into a hemispherical shape and then pressed into the object to be grasped. This causes the object to embed into the inflated membrane and maximizes A by creating both side and normal contacts. The pressure inside the membrane is then decreased, causing the membrane to tightly constrict around the object. This minimizes C during the loading phase while also activating debonding mechanisms that can increase G_c (see Fig. 2B).

To create RISOs, these soft adhesive sheets are mounted onto the surfaces of rigid robotic grippers. Our approach is not tied to any specific rigid base; RISOs can be formed by attaching the adhesive sheets to a variety of commercially available robotic grippers. In our experiments, we test a rigid Robotiq gripper⁷ as well as a Franka gripper.⁴² The placement of the adhesive sheets is also modular as follows: They can be put on the base of the gripper (Fig. 1B) or parallel to the gripper fingers (Fig. 1C). This mechanical combination and decoupling provides the fundamental advantage of our design as follows: We can have purely rigid, purely soft, or combined rigid/soft grasps depending on where the soft adhesives are mounted. However, the combination also introduces control challenges as follows: Standard control schemes for the rigid gripper may not be sufficient when the robot needs to coordinate grasps using both the rigid and soft mechanisms.

Controlling RISO grippers

RISO has the potential to enhance grasping capabilities by combining and decoupling rigid and soft mechanisms. But to harness these capabilities, we must integrate RISOs with robot arms and human operators. The key challenge here is that RISOs introduce new low-level variables (e.g., the pressure of the soft adhesives) and high-level decisions (e.g., choosing to grasp an object with the soft materials, the rigid gripper, or both). Below we formulate this control problem for RISO grippers. We then introduce solutions with varying levels of automation: from fully automated, to fully human controlled, to a shared autonomy approach between the human and robot.

Formulating the control problem

We consider settings where an RISO gripper is the end-effector of a robot arm. Let $s \in S$ be the system state, where s includes the joint position of the robot arm s_robot, the pose of the rigid gripper $s_{rigid}$ , and the pose of n soft adhesives $s_{{soft}_{1}}, \dots, s_{{soft}_{n}}$ . To pick up and manipulate objects, the robot arm must move the RISO so that either the rigid gripper $s_{rigid}$ or one of the soft adhesives $s_{{soft}_{i}}$ is in contact with the desired item. We use $o \in O$ to denote the position of objects in the environment; our robot arm observes each object o using a mounted RGB-D camera and the Yolo-v5 object detection algorithm.⁴²

There are two inputs used to move the robot arm and actuate the RISO gripper. First, the robot can take autonomous actions $a_{R} \in A$ to change its own joint velocity. Second, if a human operator is present, this human can teleoperate the robot arm and RISO gripper with a joystick (see Fig. 1A). The direction the user presses on the joystick sends a corresponding joint velocity command $a_{H} \in A$ to the robot. The objective of the system is to reach for and manipulate objects. Let $o^{*} \in O$ be the target object, and let $g^{*} \in G$ be the target grasp type. For example, if the human wants to pick up small numerous candies, then $g^{*}$ could be the soft adhesive; alternatively, if the robot arm is trying to pick up a bottle of syrup, then $g^{*}$ should be the rigid gripper. Our control problem is identifying an efficient sequence of autonomous robot actions and/or human inputs that will cause the robot to pick up object $o^{*} \in O$ using the RISO grasp type $g^{*} \in G$ .

Shared autonomy for RISO grippers

We developed three control strategies for RISO grippers along a spectrum of autonomy. At one extreme is fully autonomous control: here the robot arm acts in isolation without any human guidance. The robot leverages its RGB-D camera to detect objects in the workspace and then autonomously moves the arm with actions $a_{R}$ to align RISO with the closest object. RISO determines whether to attempt grasps using the rigid mechanism or soft adhesives based on the perceived size of the object—for objects with a height greater than 75 mm, RISO uses the rigid gripper. At the other extreme of the spectrum is fully human control: in this condition humans use a joystick to teleoperate the robot arm and RISO gripper throughout the task. The robot simply executes the human’s commanded actions $a_{H}$ to move its end-effector. When attempting to perform a soft grasp, humans must teleoperate the RISO to directly adjust low-level variables, including the normal force and gripper pressure (i.e., how hard the gripper presses down on the object and the pressure of the gripper’s pneumatic components).

Between these robot and human extremes, we propose a shared autonomy framework for RISO grippers. This approach is designed to integrate the human, robot, and RISO while reducing the human’s workload. Intuitively, under our approach the robot (1) infers the desired object $o^{*}$ and grasp $g^{*}$ from the human’s inputs and then (2) partially automates the process of reaching for and grasping that item. To estimate the human’s desired object and grasp given the previous states $s^{0 : t}$ and human actions $a_{H}^{0 : t}$ , the robot leverages Bayesian inference as follows: $P (o, g | s^{0 : t}, a_{H}^{0 : t}) = b^{t + 1} (o, g) \propto P (a_{H}^{t} | s^{t}, o, g) \cdot b^{t} (o, g)$ (2)where $P (a_{H}^{t} | s^{t}, o, g)$ models the human as a noisily rational agent,⁴³ and belief b(o, g) is the joint probability that the human wants to reach object o with grasp g. For instance, if the human teleoperates with the robot to move RISO’s soft adhesive toward a pile of candies, then b(o, g) should report that the candies are the most likely object and the soft adhesives are the most likely grasp. The robot chooses actions $a_{R}$ to assist the human in real time based on this inferred belief: $a_{R} = \sum_{o \in O} \sum_{g \in G} (o - s_{g}) \cdot b (o, g)$ (3)

In practice, Equation (3) causes the robot arm to autonomously move toward likely objects while aligning the RISO gripper for the human’s preferred grasp. This leads to a shared autonomy framework where the human is responsible for high-level decisions (e.g., which object to grasp) and the robot assists with the low-level variables (e.g., moving the robot arm directly above the object and automating the pressure changes for the soft adhesive). Our resulting shared autonomy approach is distinct from related works^44–46 because we not only need to infer the human’s goal object but also how the human wants to use the RISO to grasp that object.

Results

In this section, we conduct experiments and user studies to explore the capabilities of RISO grippers. We start by exploring how adhesion and wrapping mechanisms contribute to the soft material’s overall gripping force (“Measuring the force capacity of RISO’s soft adhesives” section) and then characterize gripping forces as a function of the target object’s radius, curvature, roughness, and porosity (“Characterizing RISO’s soft adhesives across diverse objects” section). Equipped with this understanding of the soft components, we next apply RISOs in systems with robot arms and human operators. In the “Comparing RISOs to existing grippers” section, we compare the range of objects that robots can grasp when using RISO grippers, industrial soft grippers, and a granular jamming gripper. We also conduct user studies to assess how easy it is for robots and humans to utilize RISOs with autonomous control, human control, and shared autonomy (“Reducing effort with shared autonomy” section). Put together, these four experiments characterize the soft components of RISOs, compare RISOs to state-of-the-art alternatives, and demonstrate how our controllers improve gripper performance and ease of use.

Measuring the force capacity of RISO’s soft adhesives

We start by measuring the soft adhesive’s force capacity when grasping an indenter (e.g., a target object) under controlled conditions. Figure 2C and D shows the force profiles for a 12.5 mm smooth indenter. To conduct these tests, the soft membrane was first brought into contact with the indenter, and then the membrane state was switched to a negative pressure. We finally pulled the soft gripper and indenter apart from one another until the two separated, enabling us to measure the total adhesive force F_c (see Supplementary Fig. S4). Under the positive-to-negative condition, the membrane was initially inflated (Fig. 2B), whereas for the neutral-to-negative condition, the membrane started at atmospheric pressure (Fig. 2A).

Figure 2E extends our results across indenters of different radii. To provide a baseline for these results, we included a positive-to-positive condition where the membrane was always inflated. As expected, in this positive-to-positive case, the adhesion was very low for all indenter radii; because the membrane never decreased its compliance, it was unable to achieve higher F_c values. Next, we increased the adhesion force by switching between neutral and negative states. Here we observed higher F_c as the indenter radii increased (i.e., as the contact area increased), and the soft gripper was able to reach a grasping force of 18 N for the 12.5 mm indenter. We finally repeated these same tests with the positive-to-negative condition. As before, the force capacity F_c increased with the radius of the target object, but now the grasping forces were larger because of the combination of adhesive and wrapping mechanisms (Fig. 2B). In the positive-to-negative condition, the soft adhesive reached a maximum gripping force of 50 N for an indenter radius of 12.5 mm.

Based on the adhesive scaling analysis from Equation (1), we plot F_c vs. $\sqrt{A / C}$ in Figure 2F and find a good agreement between the physical model and experimental data. For the neutral-to-negative condition, the slope gives a fracture energy of G_c = 4.2 J/m $^{2}$ , which is in line with previous results.⁴⁷ By contrast, the positive-to-negative case produced a significantly larger fracture energy of G_c = 44.7 J/m $^{2}$ . This suggests that the wrapping mechanism enhanced the capabilities of the soft adhesive by increasing G_c. We attribute this effect to the frictional sliding and hydrostatic pressure that the membrane applies to the target object when it is wrapped around that object.

Another important parameter for manipulation is the adhesion switching ratio, $S R \equiv F_{high} / F_{low}$ , where F_high is the adhesive force in the “on” state (i.e., neutral to negative or positive to negative) and F_low is the adhesive force in the “off” state (i.e., positive to positive). We calculate this for our system where the F_high is the adhesive force for the adhesion and wrapping modes, whereas the F_low is the adhesive force in the release condition. We find that the adhesion switching ratio increases with increasing indenter radii (Fig. 2G), reaching a maximum value of 187 for the 12.5 mm indenter. Overall, the soft adhesive’s ability to achieve a high force capacity (∼50 N), rapid release (<0.1 s), and high switching ratio provides key properties for object manipulation.

Characterizing RISO’s soft adhesives across diverse objects

In the “Measuring the force capacity of RISO’s soft adhesives” section, we focused on idealized objects to explore the mechanics behind the soft adhesives. We next move toward diverse objects and characterize how RISO’s soft adhesive can grasp items as a function of their curvature, roughness, and surface porosity (see Fig. 3). For each type of object, we measured the gripper’s force capacity when using the neutral-to-negative grasp (i.e., adhesion dominated), as well as the positive-to-negative grasp (i.e., combined adhesion and wrapping).

FIG. 3.

Characterizing RISO’s soft adhesives across diverse objects. (A) Curved indenters with four different curvatures. (B) Force capacity F_c vs. indenter curvature. (C) Rough indenters with lines etched at distance d from one another. (D) Force capacity F_c versus different line distances. (E) Porous indenters with four levels of porosity. (F) Force capacity F_c versus surface porosity. Across all plots, the indenters have a radius of 7.5 mm, and the scale bars are 10 mm.

We start with curved indenters in Figure 3A, B. Here the soft adhesive’s performance depended on the actuation strategy as follows: in the neutral-to-negative case, F_c decreased as the curvature increased, whereas with the positive-to-negative pressure change, F_c was roughly constant. Intuitively, we might expect the force capacity to be lower for more curved items because of a decreased surface area (i.e., the soft adhesive was only making contact with the tip of the indenter). The wrapping effect within the positive-to-negative condition mitigated this issue and enhanced the force capacity across all curvature values. We next tested target objects with engraved surfaces to create effective controlled roughness (see Fig. 3C, D). When the engraved lines were closer together, the surface was more rough—leading to lower contact area A and smaller F_c values. Conversely, when the lines were far apart, the surface was more smooth, resulting in a larger contact area and gripper forces. Here we observed a small improvement of the positive-to-negative over the neutral-to-negative condition. Finally, we measured the adhesive force when grasping objects of different surface porosity (see Fig. 3E, F). As the porosity increased the surface area decreased—this led to lower force capacity across both gripper conditions. Nonetheless, even with a surface porosity of nearly 80%, RISO’s soft gripper was able to generate adhesion forces of up to 0.9 N. RISO’s ability to grasp porous objects, as well as discrete objects of different sizes as seen in Figure 1C, contrasts with vacuum grippers. Vacuum gripper struggles to grasp and hold porous and discrete objects because they cannot maintain an airtight seal; by contrast, RISO takes advantage of its dynamic surface adhesion to grasp and release these challenging objects.

Comparing RISOs to existing grippers

Over the previous two experiments, we isolated the soft component of RISO grippers under idealized conditions. Moving forward, we will unify these soft materials with rigid end-effectors to form RISO grippers. Our core insight is that RISOs combine and decouple the rigid and soft gripper mechanisms, enabling grasps along a spectrum from purely rigid, to purely soft, to rigid–soft. We hypothesize that this decoupling will fundamentally increase the range of objects that RISOs can pick up, hold, and release. To test this hypothesis, here we explore the range of objects that robot arms and human operators can grasp with RISO grippers and state-of-the-art alternatives.

Independent variables

We compared three different grippers as follows: RISO, an industrial SoftGripper,⁴⁸ and a granular jamming gripper¹⁵ (see Fig. 4B and Supplementary Fig. S5A). Each of these grippers was mounted at the end-effector of a 7-DoF FrankaEmika robot arm.⁴⁹

FIG. 4.

Comparing RISOs to existing grippers. (A) Experimental setup. Grippers were attached to a 7-DoF robot arm and used to grasp, move, and drop a dataset of 15 household objects. (B) We compared an industrial SoftGripper, a granular jamming gripper (Granular), and RISO. (C) Success rates for each gripper when the system was fully automated. (D) Success rates for each gripper with a human-in-the-loop. A total of 12 participants remotely controlled the robot arm and grippers using a joystick. (E) After working with each gripper users responded to a 7-point Likert scale survey. Users indicate how easy it was to use the gripper and which grippers they preferred. (F) Success rates for five sample objects where RISO outperformed SoftGripper and Granular.

To understand how these different grippers performed in isolation—without any human guidance—we applied our autonomous controller from the “Controlling RISO grippers” section. Here the robot arm identified objects with a camera and then autonomously moved its gripper to align with the detected object. Next, to test how the grippers performed with a human-in-the-loop, we applied the human controller from the “Controlling RISO grippers” section. Under this approach, humans operated a joystick to remotely control the robot arm and attached grippers without any autonomous assistance. We recruited 12 participants (4 females, average age 22.9 ± 2.73) from the Virginia Tech community to take part in this study. All participants provided informed written consent following university guidelines (IRB #22-308). The order in which participants used the grippers was randomized as follows: some started with RISO, others started with a SoftGripper, and others started with the granular jamming gripper.

Dependent variables

The gripper, robot arm, and human operator collaborated to pick up, carry, and then drop a dataset of 15 household objects (see Fig. 4A and Supplementary S5B). During each interaction, the system manipulated one item as follows: The interaction was considered a success if the gripper picked up an object, carried it across the table, and dropped it in the bin. To measure the user’s perception of each gripper, we also conducted a 7-point Likert scale survey within the human-in-the-loop condition. Our questions focused on how easy it was to use the gripper and whether participants preferred one gripper over the alternatives (see Fig. 4E and Supplementary Table S1).

Autonomous results

Figure 4C shows the results of the autonomous test. This experiment compared the performance of RISO, the SoftGripper, and a granular jamming gripper when the robot’s behavior was fully automated. Across all objects and 10 autonomous trials, the manipulation success rates were SoftGripper: 56%, Granular: 51%, and RISO: 93% (see Supplementary Table S2). These differences in performance were statistically significant. A repeated measures ANOVA revealed that the gripper type had a significant effect on the overall success rate across 10 trials [ $F (2, 18) = 156.42$ , p < 0.05], and post hoc tests in Figure 4C confirmed that RISO had a higher success rate than either the pneumatic SoftGripper (p < 0.05) or the granular jamming gripper (p < 0.05).

Human-in-the-loop results

Figure 4D summarizes our results across 12 human participants. This experiment compared the performance of RISO to state-of-the-art alternatives when the system was fully controlled by human operators. We again found that the gripper type had a significant effect on the grasping success rate [ $F (2, 22) = 138.37$ , p < 0.05]. Participants successfully grasped and manipulated objects more often with RISO compared with the pneumatic SoftGripper (p < 0.05) and the granular jamming gripper (p < 0.05). Across 12 users, the success rates were SoftGripper: 58%, Granular: 41%, and RISO: 88% (see Supplementary Table S4). When surveyed using a questionnaire of Likert scale items, participants also expressed a subjective preference for the RISO gripper (see Fig. 4E and Supplementary Table S3). Participants rated the RISO gripper as easier to use than the granular jamming gripper (p < 0.05) and indicated that they would rather interact with the RISO gripper again in the future instead of either the pneumatic SoftGripper (p < 0.05) or granular jamming gripper (p < 0.05).

Comparing grippers

Whether the grippers were controlled directly by humans or autonomously by the robot arm, RISO was able to grasp, carry, and drop a more diverse range of items than the SoftGripper or Granular jamming gripper. In Figure 4F, we break down these results for some sample objects in the dataset (see all object success rates in Supplementary Table S2). Across both human and autonomous control, the pneumatic SoftGripper failed to pick up small numerous items (e.g., the nuts) or items that slip through the soft fingers (e.g., the quarter). Similarly, the granular jamming gripper struggles to grasp flat objects that the gripper cannot wrap around (e.g., the plate). RISO is able to overcome these limitations through its decoupled design as follows: We can use the soft adhesive sheets to pick up small numerous items and the rigid end-effector to grasp large heavy objects.

Reducing effort with shared autonomy

When comparing RISO to state-of-the-art grippers in the “Comparing RISOs to existing grippers” section, we tested the extremes of our controls spectrum as follows: fully autonomous or fully human controlled. Both of these control modes have advantages. On the one hand, an autonomous robot can optimize the low-level variables introduced by RISO grippers (e.g., the normal force and pneumatic pressure). On the other hand, a human can specify high-level targets and grasps (e.g., selecting whether to use the rigid, soft, or rigid–soft gripping mechanism). In the “Controlling RISO grippers” section, we combined the benefits of both approaches under a shared autonomy method for RISO grippers. This method integrates the human, robot arm, and RISO gripper as follows: Based on the human’s joystick inputs, the robot and RISO infer the human’s desired object and then automate the arm and gripper behavior to help complete the task.

Independent variables

We compared our proposed shared autonomy method to systems completely controlled by the human. Twelve new participants from the Virginia Tech community (2 females, average age 24.1 ± 4.5 years) provided informed consent and took part in this user study (IRB $# 22$ −308). These participants interacted with a joystick to remotely control the 7-DoF FrankaEmika robot arm⁴⁹ and its attached RISO gripper. Within the human condition, the robot directly executed the user’s commands without providing any assistance. By contrast, under shared autonomy, the robot and RISO provided partial assistance based on the inferred object and grasp type. The order of these conditions was counterbalanced as follows: half of the users started with shared autonomy, and the other half started with human control.

Dependent variables

Participants operated the robot and RISO to pick up, carry, and then drop the same 15 household items as in the “Comparing RISOs to existing grippers” section (also see Supplementary Fig. S5B). Our underlying hypothesis was that shared autonomy would make it easier for humans to leverage the RISO gripper. To measure changes in efficiency, we recorded the amount of time users actively controlled the robot with joystick inputs and the total length of the robot’s trajectory. Grippers that are easier to use should complete manipulation tasks with as little human guidance as necessary (e.g., fewer joystick inputs and shorter trajectory length). We also administered a 7-point Likert scale survey to assess the participants’ subjective outcomes. Our questions are listed in Supplementary Table S3 and ask about how helpful, easy to use, and intuitive the system was.

Shared autonomy results

In Figure 5B and C, we present the results of this user study. Paired t-tests revealed that the overall grasping success rates were not significantly different for either human control or shared autonomy ( $t (11) = - 0.62$ , p = 0.55, also see Supplementary Table S4). The main difference between conditions was the amount of human effort. When working with shared autonomy, participants were able to complete the manipulation tasks with fewer joystick inputs ( $t (11) = 13.65$ , p < 0.05), and the shared autonomy system moved more efficiently to complete each grasp ( $t (11) = 4.58$ , p < 0.05). This reduced the average time taken to manipulate objects: without shared control, users needed 28.4 s to grasp each item, and with shared control, this average dropped to 24.7 s. In support of these objective results, users indicated a subjective preference for working with RISO under the shared autonomy framework (Fig. 5C and Supplementary Table S3). Post hoc tests show that participants perceived the shared autonomy approach to be more helpful (p < 0.05), easier to use (p < 0.05), and overall preferable to the human controller (p < 0.05). Viewed together, these objective and subjective results suggest that RISO grippers are not only capable of picking up a diverse range of objects but also that we can reduce the human’s effort when leveraging these RISO grippers.

FIG. 5.

Making it easier for humans to utilize RISOs. (A) Overview of human control and shared autonomy. In human control, the user teleoperates with the robot and RISO throughout the entire manipulation task. By contrast, in shared autonomy, the system uses the human’s inputs to infer their desired object and grasp type. The system then partially automates the robot arm and RISO to help complete that grasp. (B) Objective results from a user study with 12 participants. With shared autonomy, users were able to complete grasps with fewer joystick inputs and shorter robot trajectories. (C) Users responded to a 7-point Likert scale survey to indicate how helpful the controller was, how easy it was to use the gripper, and which control approach they prefer to use. Higher scores indicate agreement (e.g., more helpful), and an * denotes statistical significance.

Discussion and Conclusion

In this work, we covered the surfaces of rigid end-effectors with a novel class of soft adhesives to form RISOs: RIgid–SOft grippers. Our underlying hypothesis was that RISOs would enhance grasping capabilities by both combining and decoupling rigid and soft mechanisms. Because the rigid and soft components were combined, RISO could perform rigid–soft manipulation, and because the components were decoupled, RISO could also perform purely rigid or purely soft grasps. We experimentally found that this combination enabled robots to overcome the fundamental limitations of each individual component. In our user studies, robot arms and human operators successfully utilized RISOs to pick up, move, and release a wide range of objects as follows: from flat items to 3D objects, from single items to multiple objects, and from small items to large objects. These enhanced grasping capabilities are practically useful in settings such as food processing, parts manufacturing, and assistive caregiving. We demonstrate one practical application in Figure 6, where a human operator utilized our RISO gripper to assemble a pizza. This required grasping large flat objects (the crust), manipulating tools (spreading the sauce and cheese), and picking up, transporting, and releasing small irregular food items (the pepperoni, peppers, and olives). Because RISO can perform grasps along a spectrum from purely rigid to purely soft, the robot completed each of these diverse manipulation tasks with a single gripper. Moreover, we have grasped a 2.9 kg plastic cooking oil bottle with the soft adhesive (Supplementary Fig. S7). Overall, we have demonstrated that RISOs provide exceptional gripping range to rapidly manipulate small lightweight objects (2 mg) to heavy complex items (2.9 kg) (over a 1.5 million times range in mass) and that robots and humans can effectively utilize these grippers in unstructured environments (with higher grasping success and user ratings compared with state-of-the-art alternatives).

FIG. 6.

RISO grasps and manipulates multiple food items to assemble a pizza. A human teleoperated the robot arm and attached RISO gripper. Here we show snapshots of the resulting task from a side view (top row) and an overhead view (bottom row). The RISO used purely rigid grasps to manipulate the crust, spread the sauce, and pour the cheese. For small and numerous toppings, the RISO used purely soft grasps: picking up, transporting, and releasing the pepperoni, peppers, and olives with the soft materials. See Supplementary Video for the overall task.

Another outcome of our studies was that the control scheme impacts the objective and subjective performance of the RISO gripper. This demonstrates that the way the human and robot are integrated with the gripper can have an effect on gripper outcomes. In particular, adhesion effectiveness is often characterized in controlled environments (i.e., static loading with precisely controlled alignment on testing machines) and considered to be a function of the adhesive alone.^38,50,51 However, when the adhesive or gripper is integrated with a robot arm and human user, this system now experiences variable inputs and motions that can significantly impact the functional adhesion capabilities. From the soft material viewpoint, this signifies the importance of designing in tolerance to misalignment and applied forces, consideration of adhesion switching speed and the ability to maintain high forces during the full object manipulation process, and integration of soft adhesives into larger robotic systems.

Footnotes

Data Availability

Code associated with this study is available in the . All remaining data are available upon request.

Authors’ Contributions

Conceptualization: D.P.L. and M.D.B. Data curation: M.K. and Y.K. Formal analysis: M.K., Y.K., S.A.M., J.H., R.R.S., E.P., and C.M. Funding Acquisition: D.P.L. and M.D.B. Supervision: D.P.L. and M.D.B. Visualization: M.K., Y.K., S.A.M., J.H., R.R.S., E.P., C.M., D.P.L., and M.D.B. Writing—draft/review/editing: M.K., Y.K., S.A.M., J.H., R.R.S., C.M., D.P.L., and M.D.B.

Author Disclosure Statement

All authors declare that they have no competing interests.

Funding Information

This work was supported, in part, by the National Science Foundation (Grant #2205241).

Supplementary Material

References

Sanneman

, Fourie

, Shah

. The state of industrial robotics: Emerging technologies, challenges, and key research directions. FNT in Robotics, 2021; 8(3):225–306.

Argall

. Autonomy in rehabilitation robotics: An intersection. Annu Rev Control Robot Auton Syst, 2018; 1:441–463.

Shintake

, Cacucciolo

, Floreano

, et al. Soft robotic grippers. Advanced Materials, 2018; 30(29).

Zhang

, Xie

, Zhou

, et al. State-of-the-art robotic grippers, grasping and control strategies, as well as their applications in agricultural robots: A review. Computers and Electronics in Agriculture, 2020; 177:105694.

Piazza

, Grioli

, Catalano

, et al. Catalano, and Bicchi, “A century of robotic hands. Annu Rev Control Robot Auton Syst, 2019; 2(1):1–32.

Birglen

, Schlicht

. A statistical review of industrial robotic grippers. Robotics and Computer-Integrated Manufacturing, 2018; 49:88–97.

“Products: Grippers, Camera and Force Torque Sensors-Robotiq,”. Available from: https://robotiq.com/products

Rojas

, Ma

, Dollar

. The GR2 gripper: An underactuated hand for open-loop in-hand planar manipulation. IEEE Trans Robot, 2016; 32(3):763–770.

Webb

, Chanrungmaneekul

, Yuan

, et al. Wearable roller rings to enable robot dexterous in-hand manipulation through active surfaces. arXiv Preprint arXiv:2403.13132, 2024.

10.

Ruotolo

, Brouwer

, Cutkosky

. From grasping to manipulation with gecko-inspired adhesives on a multifinger gripper. Sci Robot, 2021; 6(61):eabi9773.

11.

Glick

, Suresh

, Ruffatto

, et al. A soft robotic gripper with gecko-inspired adhesive. IEEE Robot Autom Lett, 2018; 3(2):903–910.

12.

Song

, Majidi

, Sitti

. and “Geckogripper: A soft, inflatable robotic gripper using gecko-inspired elastomer micro-fiber adhesives,” In IEEE/RSJ International Conference on Intelligent Robots and Systems, 2014, pp. 4624–4629.

13.

Tian

, Li

, Shao

, et al. Gecko-effect inspired soft gripper with high and switchable adhesion for rough surfaces. Adv Materials Inter, 2019; 6(18):1900875.

14.

Hao

, Biswas

, Hawkes

, et al. A multimodal, enveloping soft gripper: Shape conformation, bioinspired adhesion, and expansion-driven suction. IEEE Transactions on Robotics, no, 2021; 37(2):350–362.

15.

Brown

, Rodenberg

, Amend

, et al. Universal robotic gripper based on the jamming of granular material. Proc Natl Acad Sci USA, 2010; 107(44):18809–18814.

16.

Wei

, Chen

, Ren

, et al. A novel, variable stiffness robotic gripper based on integrated soft actuating and particle jamming. Soft Robotics, 2016; 3(3):134–143.

17.

Cacucciolo

, Shea

, Carbone

. Peeling in electroadhesion soft grippers. Extreme Mechanics Letters, 2022; 50:101529.

18.

Shintake

, Rosset

, Schubert

, et al. Versatile soft grippers with intrinsic electroadhesion based on multifunctional polymer actuators. Adv Mater, 2016; 28(2):231–238.

19.

Alizadehyazdi

, Bonthron

, Spenko

. An electrostatic/gecko-inspired adhesives soft robotic gripper. IEEE Robot Autom Lett, 2020; 5(3):4679–4686.

20.

Tatari

, Mohammadi Nasab

, Turner

, et al. Dynamically tunable dry adhesion via subsurface stiffness modulation. Adv Materials Inter, 2018; 5(18):1800321.

21.

Barron

III , Williams

, Tutika

, et al. A unified understanding of magnetorheological elastomers for rapid and extreme stiffness tuning. RSC Appl Polym, 2023; 1(2):315–324.

22.

Wan

, Tang

, Turner

, et al. Tunable dry adhesion of soft hollow pillars through sidewall buckling under low pressure. Adv Funct Materials, 2023; 33(2):2209905.

23.

Song

, Sitti

. Soft grippers using micro-fibrillar adhesives for transfer printing. Adv Mater, 2014; 26(28):4901–4906.

24.

Song

, Drotlef

D-M

, Majidi

, et al. Controllable load sharing for soft adhesive interfaces on three-dimensional surfaces. Proc Natl Acad Sci U S A, 2017; 114(22):E4344–E4353.

25.

Coulson

, Stabile

, Turner

, et al. Versatile soft robot gripper enabled by stiffness and adhesion tuning via thermoplastic composite. Soft Robot, 2022; 9(2):189–200.

26.

, Dong

, Sun

. Soft gripper design based on the integration of flat dry adhesive, soft actuator, and microspine. IEEE Trans Robot, 2021; 37(4):1065–1080.

27.

Chen

, Zhu

, Cao

, et al. Soft-rigid coupling grippers: Collaboration strategies and integrated fabrication methods. Sci China Technol Sci, 2023; 66(11):3051–3069.

28.

Hernandez

, Sunny

MSH

, Sanjuan

, et al. Current designs of robotic arm grippers: A comprehensive systematic review. Robotics, 2023; 12(1):5.

29.

Nasab

, Sabzehzar

, Tatari

, et al. A soft gripper with rigidity tunable elastomer strips as ligaments. Soft Robot, 2017; 4(4):411–420.

30.

Park

, Seo

, Bae

. A hybrid gripper with soft material and rigid structures. IEEE Robot Autom Lett, 2019; 4(1):65–72.

31.

, Xie

, Wang

, et al. Stiffness-tunable soft gripper with soft-rigid hybrid actuation for versatile manipulations. Soft Robot, 2022; 9(6):1108–1119.

32.

Peng

, Liu

, Song

, et al. The enhanced adaptive grasping of a soft robotic gripper using rigid supports. ASI, 2024; 7(1):15.

33.

Gafer

, Heymans

, Prattichizzo

, et al. and “The quad-spatula gripper: A novel soft-rigid gripper for food handling,” In IEEE International Conference on Soft Robotics, 2020, 39–45.

34.

Keely

, Nemlekar andH, Losey

. Kiri-Spoon: A soft shape-changing utensil for robot-assisted feeding. arXiv Preprint arXiv:2403.05784, 2024.

35.

Guo

X-Y

, Li

W-B

, Gao

Q-H

, et al. Self-locking mechanism for variable stiffness rigid–soft gripper. Smart Mater Struct, 2020; 29(3):35033.

36.

Mehta

, Kim

, Hoegerman

, et al. and “Riso: Combining rigid grippers with soft switchable adhesives,” in IEEE International Conference on Soft Robotics, 2023.

37.

Bartlett

, Croll

, King

, et al. Looking beyond fibrillar features to scale gecko-like adhesion. Adv Mater, 2012; 24(8):1078–1083.

38.

Croll

, Hosseini

, Bartlett

. Switchable adhesives for multifunctional interfaces. Adv Materials Technologies, 2019; 4(8).

39.

Autumn

, Sitti

, Liang

, et al. Evidence for van der Waals adhesion in gecko setae. Proc Natl Acad Sci U S A, 2002; 99(19):12252–12256.

40.

Lee

, Shi

, Jung

, et al. Bioinspired materials for underwater adhesion with pathways to switchability. Cell Reports Physical Science, 2023; 4(10):101597.

41.

Bartlett

, Case

, Kinloch

, et al. Peel tests for quantifying adhesion and toughness: A review. Progress in Materials Science, 2023; 137:101086.

42.

Jiang

, Ergu

, Liu

, et al. A review of Yolo algorithm developments. Procedia Computer Science, 2022; 199:1066–1073.

43.

Jeon

, Milli

, Dragan

. Reward-rational (implicit) choice: A unifying formalism for reward learning. Advances in Neural Information Processing Systems, 2020:4415–4426.

44.

Jain

, Argall

. Probabilistic human intent recognition for shared autonomy in assistive robotics. ACM Trans Hum Robot Interact, 2019; 9(1):1–23.

45.

Javdani

, Admoni

, Pellegrinelli

, et al. Shared autonomy via hindsight optimization for teleoperation and teaming. The International Journal of Robotics Research, 2018; 37(7):717–742.

46.

Losey

, Jeon

, Li

, et al. Learning latent actions to control assistive robots. Auton Robots, 2022; 46(1):115–147.

47.

Swift

, Haverkamp

, Stabile

, et al. Active membranes on rigidity tunable foundations for programmable, rapidly switchable adhesion. Adv Materials Technologies, 2020; 5(11):2000676.

48.

“Dobot Magician SoftGripper - 3 Fingers,”. Available from: https://www.soft-gripping.shop/en/dobot-magician-softgripper-3-fingers.html

49.

“Franka Emika - Next Generation Robotics,”. Available from: https://www.franka.de/

50.

Jagota

, Hui

C-Y

. Adhesion, friction, and compliance of bio-mimetic and bio-inspired structured interfaces. Materials Science and Engineering: R: Reports, 2011; 72(12):253–292.

51.

Creton

, Ciccotti

. Fracture and adhesion of soft materials: A review. Rep Prog Phys, 2016; 79(4):46601.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

5.99 MB

4.04 MB

1.56 MB

2.30 MB

0.04 MB

6.47 MB

4.48 MB

0.51 MB

10.25 MB

0.03 MB

0.04 MB

0.03 MB

0.00 MB