SHaRe-RL: Structured, Interactive Reinforcement Learning for Contact-Rich Industrial Assembly Tasks
Jannick Stranghöner, Philipp Hartmann, Lisa-Marie Weigelt, Marco Braun, Sebastian Wrede, Klaus Neumann
AI summary
Problem
High-mix low-volume industrial assembly demands precision and flexibility, but existing robotic systems fail due to brittle programming and the sample inefficiency and safety risks of contact-rich reinforcement learning.
Approach
SHaRe-RL embeds interactive off-policy RL within a sequence of manipulation primitives, guided by human demonstrations and bounded by adaptive per-axis force limits.
Key results
- Robust connector insertion with 0.2–0.4 mm clearance within 3 hours of real-world training
- Outperforms unstructured human-in-the-loop RL and matches skilled human cycle times
- Generalizes to previously unseen connector variants without retraining
- Proves adaptive force limits safely bound contact forces while preserving free-space dynamics
Why it matters
Enables safe, flexible, and economically viable deployment of reinforcement learning for industrial automation in SMEs.
Abstract
High-mix low-volume (HMLV) industrial assem- bly, common in small and medium-sized enterprises (SMEs), re- quires the same precision, safety, and reliability as high-volume automation while remaining flexible to product variation and environmental uncertainty. Current robotic systems struggle to meet these demands. Manual programming is brittle and costly to adapt, while learning-based methods suffer from poor sample efficiency and unsafe exploration in contact-rich tasks. To address this, we present SHaRe-RL, a reinforcement learning framework that leverages multiple sources of prior knowledge. By (i) structuring skills into manipulation primitives, (ii) incorporating human demonstrations and online corrections, and (iii) bounding interaction forces with per-axis compliance, SHaRe-RL enables efficient and safe online learning for long- horizon, contact-rich industrial assembly tasks. Experiments on the insertion of industrial Harting connector modules with 0.2–0.4 mm clearance show reliable learning within practical wall-clock budget and improved performance over an unstruc- tured human-in-the-loop RL baseline. We further show that the learned policy generalizes to previously unseen connector variants. Overall, our results show that process expertise alone can effectively guide real-world RL, making deployment safer, more robust, and economically viable for industrial assem- bly. Source code and demonstration videos are available at https://share-rl.github.io/share-rl.io/