Soft Robotics: AI Grasp-Planning for Non-Dexterous Grippers

Modern robotics is moving away from rigid, multi-fingered hands toward soft, adaptable mechanisms like jam-based granular grippers and origami vacuum actuators. This article explores how these non-dexterous systems successfully manipulate complex, irregular objects by combining compliant physical materials with advanced grasp-planning artificial intelligence (AI). By offloading the mechanical complexity of finger dexterity onto intelligent software and morphological adaptability, these grippers achieve high reliability in unpredictable environments.

Jam-based granular grippers operate without any traditional joints or fingers. They consist of an elastic membrane filled with fine granular material, such as coffee grounds, glass beads, or plastic particles. In its unjammed state, the gripper is soft and malleable. When pressed against an irregular object, the membrane flows around the target, conforming to its exact shape. Once in place, a vacuum pump removes the air from inside the membrane. This vacuum causes the granules to pack tightly together, transitioning the gripper from a fluid-like state to a rigid, solid-like state through a process called jamming. The gripper locks onto the object using friction, vacuum suction, and mechanical interlocking.

Origami vacuum actuators take a structured approach to soft manipulation. Inspired by traditional paper folding, these grippers are built from flexible, pre-creased elastomeric or polymeric sheets enclosing an airtight chamber. When negative pressure (a vacuum) is applied to the internal chamber, the walls collapse neatly along the designated fold lines. This predictable buckling motion generates a strong contracting and pinching force. Even without individually controlled finger joints, the folded structure naturally wraps around uneven surfaces, distributing pressure evenly and securing odd-shaped items without crushing them.

While both of these grippers physically adapt to shapes on contact, they still require grasp-planning AI to handle irregular objects efficiently. Without fingers to adjust a poor grip mid-lift, the initial point of contact and angle of approach are critical. Grasp-planning AI uses depth cameras and 3D sensors to scan the target object and generate a point cloud. Deep learning models then analyze this spatial data to identify the object's center of mass, surface texture, and geometric features.

For granular grippers, the AI calculates the optimal trajectory and downward force needed to maximize the surface area covered by the membrane before the vacuum is triggered. If the gripper hits an object off-center, the object might slip during the jamming phase. The AI prevents this by finding the most stable, enveloping contact patch.

For origami actuators, the AI aligns the specific geometry of the folding creases with the contours of the target object. By simulating how the origami structure will deform under vacuum, the AI positions the actuator so its collapsing walls lock under the object's ledges or curve around its widest points.

By combining the physical intelligence of soft materials with the computational power of grasp-planning AI, robotic systems no longer need complex, fragile finger dexterity. The AI guarantees the gripper approaches the object at the ideal location and angle, while the granular jamming or origami folding physics handle the micro-adjustments needed for a secure hold.