README.md

Awesome-multi-step-manipulation

Inspired by the awesome list and awesome-long-horizon-goal-reaching.

Reading list by Junyeob Baek , with help from Haegu Lee.

This repository is one of project deliverables for project-sandwich-man, which project is especially focused on hierarchical stacking blocks. If you see papers missing from the list, please send me an email or a pull request.

Table of Contents

• Awesome-block-stacking-manipulation
  • Table of Contents
  • Core Area
    • Approach from TAMP
    • Approach from Hierarchical RL
    • Multi-step Manipulation
    • Multi-goal Task Approach
    • Approach from Relational RL
    • Good Environment to Start
    • Further More

Core Area

Approach from TAMP

• Classical TAMP vs Learning-based TAMP
• Classical TAMP

Approach from Hierarchical RL

• nikhilbarhate99/Hierarchical-Actor-Critic-HAC-PyTorch
• Hierarchical Deep Reinforcement Learning (HDQN)
• Learning Multi-Level Hierarchies with Hindsight(HAC)
• Introduce to HRL
• Meta-Learning shared Hierarchy(MLSH)
• Article : The Promise of Hierarchical Reinforcement Learning

Multi-step Manipulation

(for long-horizon tasks)

• A Long Horizon Planning Framework for Manipulating Rigid Pointcloud Objects
  • its manipulation skills from here
• Hierarchical Reinforcement Learning with Universal Policies for Multi-Step Robotic Manipulation, video
• Planning for Multi-Stage Forceful Manipulation

Multi-goal Task Approach

• Hindsight Experience Replay
• CHER: Curriculum-guided Hindsight Experience Replay , code: https://github.com/mengf1/CHER

Exploration by Curiosity

• CURIOUS: Intrinsically Motivated Modular Multi-Goal Reinforcement Learning
• Curious hierarchical actor-critic reinforcement learning

Approach from Relational RL

• Towards Practical Multi-Object Manipulation using Relational Reinforcement Learning

Good Environment to Start

• IanYangChina/pybullet_multigoal_gym, paper : https://arxiv.org/pdf/2105.05985.pdf
• qgallouedec/panda-gym -> pre-trained model on rl-baselines3-zoo

---

Further More

• IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks, (https://clvrai.github.io/furniture)
• Learning to guide task and motion planning using score-space representation
• Self-play for goal discovery in robotics manipulation
• Transporter Network: Rearranging the Visual World for Robotic Manipulation
• CAVIN: Dynamics Learning with Cascaded Variational Inference for Multi-Step Manipulation
• Neural Task Programming: Learning to Generalize Across Hierarchical Tasks