TacEvo: Self-Evolving Architecture Discovery for Robotic Tactile Perception via LLM-Driven Quality-Diversity Search

2026-06-29

AI Summary

This work addresses the limitations of conventional tactile perception networks, which rely on handcrafted designs, and existing neural architecture search methods that suffer from high computational costs and restricted structural diversity. The authors propose a self-evolving architecture discovery framework that, for the first time, integrates large language model (LLM)-driven code-level mutation and crossover operations into a quality–diversity optimization loop. By introducing “architectural diversity” and “efficiency ratio” as behavioral descriptors, the framework autonomously evolves novel and efficient tactile perception architectures. Evaluated on the ViTacTip dataset, the generated architectures achieve trainability rates of 96.0% and 94.5%, with validation performance improvements of 56.1% and 96.1% within 20 generations. High-fidelity assessments further demonstrate force prediction accuracy comparable to expert-designed models and significantly superior fine-grained grating classification over current baselines.

Abstract

Vision-based tactile sensing converts contact-induced surface deformation into images, enabling robots to infer contact forces and fine surface textures that are not accessible through conventional vision alone. However, tactile images are sensor- and physics-specific, so effective architectures often require expert intuition and extensive manual iteration. Existing neural architecture search (NAS) pipelines can reduce this burden, but they are often computationally expensive and restricted to hand-designed search spaces, which limits architectural novelty and diversity. We introduce TacEvo, a self-evolving architecture discovery framework that improves network designs from downstream feedback. TacEvo uses an LLM to generate code-level mutations and crossovers, and a MAP-Elites quality-diversity loop that preserves diverse elite architectures while preferentially reusing prompts that consistently yield improvements. Exploration is guided by two behavioural descriptors, Architectural Diversity and Efficiency Ratio, which encourage coverage across structural variations and compute-size trade-offs. On ViTacTip force regression and grating classification, TacEvo achieves high autonomous generation reliability (96.0%/94.5% trainable) and improves best validation fitness over 20 generations by 56.1%/96.1%. In a 20-seed post-search high-fidelity evaluation, TacEvo matches the expert baseline on force prediction and outperforms it on fine-grained grating classification. These results suggest that LLM-driven self-evolving search constitutes a practical paradigm for AI-assisted scientific discovery in specialised robotic sensing.

Problem

• tactile perception
• neural architecture search
• quality-diversity
• robotic sensing
• vision-based tactile sensing

Innovation

• LLM-driven architecture search
• quality-diversity optimization
• self-evolving neural networks
• tactile perception
• MAP-Elites