TaSA: Two-Phased Deep Predictive Learning of Tactile Sensory Attenuation for Improving In-Grasp Manipulation
Pranav Ponnivalavan, Satoshi Funabashi, Alexander Schmitz, Tetsuya Ogata, Shigeki Sugano
AI summary
Problem
Robotic hands struggle to distinguish self-contact from external object contact during multi-finger manipulation, leading to unstable grasps and poor generalization in contact-rich scenarios.
Approach
TaSA employs a two-phase deep learning framework that first models self-touch dynamics and then integrates these predictions into a motion-learning network to filter predictable tactile feedback and highlight external interactions.
Key results
- Accurate prediction of self-touch tactile patterns across diverse finger configurations
- Significantly higher success rates in precision insertion tasks compared to raw tactile baselines
- Effective generalization across varying object sizes, positions, and insertion angles
- Clear disambiguation of self-contact versus external object interactions in tactile streams
Why it matters
Provides a biologically inspired tactile filtering mechanism that enhances robustness and dexterity for real-world robotic manipulation.
Abstract
Humans can achieve diverse in-hand manipulations, such as object pinching and tool use, which often involve simultaneous contact between the object and multiple fingers. This is still an open issue for robotic hands because such dexterous manipulation requires distinguishing between tactile sensations generated by their self-contact and those arising from external contact. Otherwise, object/robot breakage happens due to contacts/collisions. Indeed, most approaches ignore self- contact altogether, by constraining motion to avoid/ignore self- tactile information during contact. While this reduces complexity, it also limits generalization to real-world scenarios where self-contact is inevitable. Humans overcome this challenge through self-touch perception, using predictive mechanisms that anticipate the tactile consequences of their own motion, through a principle called sensory attenuation, where the nervous system differentiates predictable self-touch signals, allowing novel object stimuli to stand out as relevant. Deriving from this, we introduce TaSA, a two-phased deep predictive learning framework. In the first phase, TaSA explicitly learns self-touch dynamics, modeling how a robot’s own actions generate tactile feedback. In the second phase, this learned model is incorporated into the motion learning phase, to emphasize object contact signals during manipulation. We evaluate TaSA on a set of insertion tasks, which demand fine tactile discrimination: inserting a pencil lead into a mechanical pencil, inserting coins into a slot, and fixing a paper clip onto a sheet of paper, with various orientations, positions, and sizes. Across all tasks, policies trained with TaSA achieve significantly higher success rates than baseline methods, demonstrating that structured tactile perception with self-touch based on sensory attenuation is critical for dexterous robotic manipulation.