Tactile Object Recognition with Recurrent Neural Networks through a Perceptive Soft Gripper
Enrico Donato, David Pelliccia, Matin Hosseinzadeh, Mahmood Amiri, Egidio Falotico
AI summary
Problem
Soft robotic grippers require reliable tactile perception to identify grasped objects, yet single-touch sensing is often ambiguous and multi-modal sensor integration remains underexplored for optimal recognition.
Approach
The authors develop a two-stage LSTM framework that first estimates object geometry to refine subsequent tactile classification, combined with an iterative grasping loop and simulation-based sensor layout optimization to maximize recognition accuracy with minimal hardware.
Key results
- Dual-stage LSTM achieves high recognition accuracy for over sixteen heterogeneous objects in both simulation and physical tests
- Multi-modal sensing (curvature and pressure) significantly outperforms single-modality approaches
- Iterative consecutive grasps progressively increase classification confidence beyond predefined thresholds
- Computational design identifies minimal optimal sensor configurations balancing performance and system complexity
Why it matters
This work provides a practical, hardware-efficient perception pipeline that enables soft robots to reliably identify objects during manipulation, advancing autonomous handling and adaptive robotics.
Abstract
Soft robot perception integrates information from distributed, multi-modal sensors, broadening their application to active interaction. Our work introduces recurrent learning models for tactile-based object recognition, demonstrating comparable performance in virtual and real-world scenarios. The work focuses on soft grippers, which facilitate adaptation to objects of varying shapes and sizes thanks to passive finger compliance. Our model successfully identifies over sixteen heterogeneous objects. Findings underscore the significance of sensory multi-modality over single. We highlight how spatial distribution and sensory signal dynamics influence overall estimation accuracy, and what the minimal grasp set is to achieve certain recognition.