Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SELF-VLA: A Skill Enhanced Agentic Vision-Language-Action Framework for Contact-Rich Disassembly

Chang Liu, Sibo Tian, Xiao Liang, Minghui Zheng

Abstract

Disassembly automation has long been pursued to address the growing demand for efficient and proper recovery of valuable components from the end-of-life (EoL) electronic products. Existing approaches have demonstrated promising and regimented performance by decomposing the disassembly process into different subtasks. However, each subtask typically requires extensive data preparation, model training, and system management. Moreover, these approaches are often task- and component-specific, making them poorly suited to handle the variability and uncertainty of EoL products and limiting their generalization capabilities. All these factors restrict the practical deployment of current robotic disassembly systems and leave them highly reliant on human labor. With the recent development of foundation models in robotics, vision-language-action (VLA) models have shown impressive performance on standard robotic manipulation tasks, but their applicability to complex, contact-rich, and long-horizon industrial practices like disassembly, which requires sequential and precise manipulation, remains limited. To address this challenge, we propose SELF-VLA, an agentic VLA framework that integrates explicit disassembly skills. Experimental studies demonstrate that our framework significantly outperforms current state-of-the-art end-to-end VLA models on two contact-rich disassembly tasks. The video illustration can be found via https://zh.engr.tamu.edu/wp-content/uploads/sites/310/2026/03/IROS-VLA-Video.mp4.

SELF-VLA: A Skill Enhanced Agentic Vision-Language-Action Framework for Contact-Rich Disassembly

Abstract

Disassembly automation has long been pursued to address the growing demand for efficient and proper recovery of valuable components from the end-of-life (EoL) electronic products. Existing approaches have demonstrated promising and regimented performance by decomposing the disassembly process into different subtasks. However, each subtask typically requires extensive data preparation, model training, and system management. Moreover, these approaches are often task- and component-specific, making them poorly suited to handle the variability and uncertainty of EoL products and limiting their generalization capabilities. All these factors restrict the practical deployment of current robotic disassembly systems and leave them highly reliant on human labor. With the recent development of foundation models in robotics, vision-language-action (VLA) models have shown impressive performance on standard robotic manipulation tasks, but their applicability to complex, contact-rich, and long-horizon industrial practices like disassembly, which requires sequential and precise manipulation, remains limited. To address this challenge, we propose SELF-VLA, an agentic VLA framework that integrates explicit disassembly skills. Experimental studies demonstrate that our framework significantly outperforms current state-of-the-art end-to-end VLA models on two contact-rich disassembly tasks. The video illustration can be found via https://zh.engr.tamu.edu/wp-content/uploads/sites/310/2026/03/IROS-VLA-Video.mp4.
Paper Structure (13 sections, 6 figures, 1 table, 1 algorithm)

This paper contains 13 sections, 6 figures, 1 table, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Robotic Disassembly
  4. Vision-Language-Action Models
  5. METHODOLOGIES
  6. SELF-VLA
  7. Dataset Construction
  8. Fine-tuning Process
  9. EXPERIMENTS AND TEST RESULTS
  10. Environmental Setup
  11. Evaluation
  12. Results
  13. CONCLUSIONS

Figures (6)

  • Figure 1: SELF-VLA framework
  • Figure 2: Comparison between current VLA approaches and our SELF-VLA framework.
  • Figure 3: The complexity of the CPU extraction task in EoL desktop disassembly.
  • Figure 4: Experimental Studies. Top figures show the results for RAM disassembly, and bottom figures demonstrate CPU disassembly case.
  • Figure 5: Experimental Study of the VLA-corrector: A human interrupts the skill execution, after which the VLA-corrector attempts to pick the CPU up again and complete the remaining task.
  • ...and 1 more figures