Autonomous selection of optimal gripping strategies for deformable objects requires combining sensing, physical modeling, and adaptive control so robots can handle variability in shape, stiffness, and surface properties. Research by Antonio Bicchi at University of Pisa has emphasized the importance of contact modeling and tactile information for predicting how soft materials respond to force. Practical tasks range from handling fresh produce to folding garments and manipulating tissue in surgery, each imposing different accuracy and safety constraints.
Sensing and modeling
Accurate perception is central. Visual sensors provide global shape and pose, while tactile arrays and force sensors offer local deformation data that reveal material properties. Daniela Rus at Massachusetts Institute of Technology has advanced soft robotics and compliance sensing that make interaction safer around humans. Model-based methods use physics simulations such as simplified finite element models to predict deformation under candidate grips, enabling anticipatory planning. Nuance arises because high-fidelity models are computationally costly and often mismatch real materials, so practitioners must balance fidelity and speed.
Planning and learning strategies
Learning-based approaches train policies that map perception to actions, bypassing explicit physical models. Sergey Levine at University of California Berkeley has demonstrated reinforcement learning and imitation techniques that generalize grips across varied items. Ken Goldberg at University of California Berkeley has explored algorithmic methods for cloth manipulation that combine perception-driven heuristics with learning to reduce failure modes. Hybrid strategies that fuse model predictions with data-driven adaptation tend to perform best in practice, using simulation for bulk training and tactile feedback for online correction. Nuance includes the sim-to-real gap where policies trained in virtual environments require domain adaptation.
Relevance stems from economic, cultural, and safety consequences. In manufacturing and textile industries robotic competence with deformables can reduce repetitive manual labor and increase production speed, while in healthcare improved manipulation of soft tissue can enhance surgical outcomes. Environmental implications appear in food handling where gentle gripping reduces waste. Conversely automation may disrupt livelihoods in regions reliant on manual garment work, making human-centered deployment and retraining important.
Robust deployment emphasizes iterative calibration, active perception to probe material properties, and safety-driven compliance control. Combining evidence-based modeling from Antonio Bicchi at University of Pisa with learning advances from Sergey Levine and algorithmic insights from Ken Goldberg at University of California Berkeley yields practical paths for robots to autonomously select optimal gripping strategies for deformable objects. Nuance remains in balancing generality, safety, and economic impacts.