Research Analyzer
← Back ICRA 2026

Tacser and Action-Conditioned Latent Filter for Generalizable Robotic Surface Perception

Anirvan Dutta, Yerkebulan Massalim, etienne burdet, Mohsen Kaboli

PDF

AI summary

Key figure (auto-extracted from paper)
An unsupervised action-conditioned deep state-space model combined with induced high-frequency vibrations enables robots to continuously and generalizably perceive surface textures without labeled data.
Tactile sensing texture perception deep state-space models unsupervised learning robotic manipulation vibro-tactile interaction

Problem

Prior tactile texture perception relies on discriminative models with fixed categories or overlooks interaction dynamics, failing to capture the continuous, causal physical properties required for adaptive robotic manipulation.

Approach

The authors introduce Tacser, a device that induces high-frequency micro-vibrations during sliding, paired with a Latent Filter—an unsupervised variational deep state-space model that infers continuous textural properties from action-conditioned tactile observations.

Key results

  • Unsupervised action-conditioned deep state-space model for continuous texture inference
  • Tacser device enabling independent high-frequency vibration during sliding
  • Latent Filter with Bayesian integration for temporally coherent latent manifolds
  • Real-robot validation showing superior generalization over state-of-the-art baselines

Why it matters

Enables robots to adaptively perceive and manipulate diverse surfaces in dynamic environments, advancing autonomous manipulation and interactive perception.

Abstract

Perceiving the physical properties of different sur- faces/textures via tactile sensing has been a long-standing prob- lem in robotics. Most prior work has been limited to dis- criminative models that classify textures into a fixed set of categories. However, to enable seamless and efficient autonomous manipulation, robots must infer physical properties as structured, continuous variables rather than as discrete class labels. In this work, we present a novel deep state-space model (DSSM) to learn and infer key causal textural properties in an unsupervised manner. Using variational inference to solve the DSSM, our proposed Latent Filter allows robotic systems to perceive textures in a continuous and generalizable manner. In addition, we explore a novel interaction approach: Tacser (Tactile Enhancer), to fur- ther enhance tactile sensing through vibrations induced by high- frequency micro-movements and thereby improve perception. We evaluated our approach against state-of-the-art techniques and performed extensive ablation studies to demonstrate its effec- tiveness. This work advances tactile-based texture perception, providing a generalizable and comprehensive framework for robotics.

Index terms

Force and Tactile Sensing Perception for Grasping and Manipulation Representation Learning

Related papers