Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
September 10, 2025Robotica

Image segmentation-driven sim-to-real deep reinforcement learning framework for accurate peg-in-hole assembly

View Full Paper
Ask AI
Bookmark
Share

Authors

NZNing ZhangYZYongjia ZhaoMYMinghao Yang

Discussion

Loading...

Member takes

Overview

This framework demonstrates improved robotic assembly success by addressing the simulation-to-reality gap, highlighting adaptability in complex environments.

Key Points

  • The proposed method improves peg-in-hole assembly success by 75%, overcoming traditional programming limitations.
  • It employs a U-net segmentation model, achieving 15 successful assembly tests out of 20 in real-world applications.
  • Integration of visual and force feedback enhances precision in insertion tasks across varying light conditions.
  • Evaluation includes simulations and real robotic environments, indicating robust performance despite challenges.

Cite This Study

Zhang et al. (2025) studied this question.

synapsesocial.com/papers/68c1c22554b1d3bfb60ef1ebhttps://doi.org/10.1017/s026357472510177x
View Full Paper
Ask AI
Bookmark
Share