Home

Central Wiki for Large-Scale and Efficient Bin-Picking

This project aims to render a work like QT-Opt sample efficient by using invariance and equivariance principles like those in ITER with a large set of diverse and complex objects.

Table of Contents

• Python Resources
• Software and Hardware
• Core Research Papers
• Development Stages
  • Simulation:
    • Objects: Object categorization, generation, inclusion, and arrangement.
  • State Representation:
    • Graphs
  • Real-Robot Experiments:
    • ROS-Interfaces in Robosuite

---

Python

• Resources: python, numpy, scipy, matplotlib, conda, virtualenv

---

Software and Hardware

• Modelling: Solidworks
• Simulator: Mujoco 2.0
• Toolsuite: Robosuite.ai Home Page Robosuite Code
• Code-base: Robosuite-Benchmarks built on top of rlkit
• Graph Based Code: Relational RL built on top of rlkit
• Object Detection and Tracking: TBD
• Graph Lib: Deep Graph Lib
• Panda Robot: [Specs] [ROS-panda] [FCI] [ROS-Panda Interfaces adapted for DRL]
• Camera: TBD
• Objects: TBD

---

Core Research Papers

• Large-Scale Bin-Picking

  • QT-Opt [Project Page] [Paper] [Model Code]
  • Quantile QT-OPT: [Project]
  • Never Stop Learning: [Project Page] [Vid1 Vid2] [Paper]

• Invariance

  • ITER: [Project] [Video] [Paper] [Code]
  • Meta-Learning Symmetries: [Paper] [Code]
  • Reinforcement Learning with Augmented Data: [Paper] [Video] [Code]

• Graphs

  • Towards Practical Multi-Object Manipulation using Relational Reinforcement Learning. Proceedings. [Project] [Paper] Video] [Code]
  • Visual-Semantic Graph Attention Network for Human-Object Interaction Detection: [Project] [Paper] [Code] [Video]
  • Magnet: Deep Multi-Agent Reinforcement Learning with Relevance Graphs: [Paper] [Code]
  • Deep Reinforcement Learning meets Graph Neural Networks-exploring a routing optimization use case: [Paper] [Code]

---