PhD-manipulation

Learning-based slip detection for adaptive grasp control in robotic manipulation

PhD Thesis Project — Théo Ayral
CEA (Leti & List) · Université Paris-Saclay

CONTEXT: TraceBot Project

Robotic manipulation in critical environements.

TraceBot is an EU project developing robotic systems that can verify and trace manipulation actions (“audit trail”) for regulated environments, with a focus on handling sterile medical products.

TraceBot / manipulation platform

Hardware

Multifingered gripper & actuation

Four-fingered, 14-DoF gripper
Backdrivable tendon actuation enabling haptic interaction
Designed for multi-contact grasping and internal force control

Piezoelectric tactile sensing (PzE) — slip detection

Piezoelectric sensors capture high-frequency dynamics, enabling detection of dynamic tactile events and incipient slip.

Piezoelectric PVDF–TrFE tactile transducers
High-bandwidth sensing: 10 kHz sampling (30 Hz – 2.5 kHz effective)
Captures friction-induced vibrations and dynamic slip events

→ Project page for slip detection using spectro-temporal learning

Piezoresistive tactile arrays (PzR) — contact estimation

Piezoresistive arrays provide pressure distribution and contact localization, supporting grasp modeling and internal force coordination.

Piezoresistive polymer (Velostat™) tactile arrays
8×8 taxel matrix from orthogonal electrode sheets
Provides pressure distribution, contact area, and CoP estimation

→ Project page for internal-force control

Contributions

The contributions build on each other, following the narrative flow of the thesis:

Thesis contribution flow: slip detection → robustness → reactive slip control

C1 — Early slip detection from tactile vibrations

Detect incipient slip from high-bandwidth tactile vibrations using learning-based spectro-temporal analysis, in real time (100 Hz).

Related publications

AIM 2023 (published)
Spectro-Temporal Recurrent Neural Network for Robotic Slip Detection with Piezoelectric Tactile Sensor
Théo Ayral, Saifeddine Aloui, Mathieu Grossard

Project page

→ project page: https://thayral.github.io/tactile-slip-detection-pze/
→ GitHub demo: https://github.com/thayral/tactile-slip-demo

C2 — Data-driven robustness to manipulation perturbations

Improve robustness to transient events and actuation noise through perturbation-aware training and haptic data fusion, reducing false alarms while preserving low detection latency.

Robustness: 38.77 % → 90.43 %
Average detection delay: 24.1 ms
Perfect recall on slip events

Related publications

(in preparation)
Robust Tactile Slip Detection under Manipulation Perturbations Théo Ayral, Saifeddine Aloui, Mathieu Grossard

C3 — Closed-loop adaptation of grasp forces

Stabilize multi-fingered grasps by injecting internal forces based on tactile feedback, without relying on explicit friction models.

Related publications

ICRA 2026 (accepted)
Reactive Slip Control in Multifingered Grasping: Hybrid Tactile Sensing and Internal-Force Optimization
Théo Ayral, Saifeddine Aloui, Mathieu Grossard
https://arxiv.org/abs/2602.16127
Patent application (2025)
Robotic gripper and control method
M. Grossard, S. Aloui, T. Ayral — US Patent Application 19/011,931

Project page

→ Reactive slip control:
https://thayral.github.io/reactive-slip-control

Code & Resources

Demo (minimal runnable example): https://github.com/thayral/tactile-slip-demo
PhD defense slides: https://thayral.github.io/phd-defense-slides/
Thesis manuscript (PDF): https://theses.hal.science/tel-05468226

Contact

Théo AYRAL