Door-opening task
The door-opening task is decomposed into reaching, grasping, turning, and pulling, where all tasks except grasping are trained using SAC(Soft Actor-Critic) and performed sequentially.
We provide a method for task decomposition and task-specific rewards for high-level tasks.
This paper introduces a reinforcement learning method that leverages task decomposition and a task-specific reward system to address complex high-level tasks, such as door opening, block stacking, and nut assembly. These tasks are decomposed into various subtasks, with the grasping and putting tasks executed through single joint and gripper actions, while other tasks are trained using the SAC algorithm alongside the task-specific reward system. The task-specific reward system aims to increase the learning speed, enhance the success rate, and enable more efficient task execution. The experimental results demonstrate the efficacy of the proposed method, achieving success rates of 99.9% for door opening, 95.25% for block stacking, 80.8% for square-nut assembly, and 90.9% for round-nut assembly. Overall, this method presents a promising solution to address the challenges associated with complex tasks, offering improvements over the traditional end-to-end approach.
The door-opening task is decomposed into reaching, grasping, turning, and pulling, where all tasks except grasping are trained using SAC(Soft Actor-Critic) and performed sequentially.
The block-stacking task is decomposed into reaching, grasping, reaching, and putting, where all tasks except grasping and putting are trained using SAC and performed sequentially.
The nut-assembly task is decomposed into reaching, aligning, reaching, grasping, assembly, and putting, where all tasks except grasping and putting are trained using SAC and performed sequentially.
The nut-assembly task is decomposed into reaching, aligning, reaching, grasping, assembly, and putting, where all tasks except grasping and putting are trained using SAC and performed sequentially.
There's a lot of excellent work that was introduced around the same time as ours.
The Task Decomposition and Dedicated Reward-System-Based Reinforcement Learning Algorithm for Pick-and-Place serves as the foundation of the methodology outlined in the paper 'Reinforcement Learning with Task Decomposition and Task-Specific Reward System for Automation of High-Level Tasks.
Towards Hierarchical Task Decomposition using Deep Reinforcement Learning for Pick and Place Subtasks introduces pick-and-place divided into approach, manipulate, and retract, and successfully performs pick-and-place using DDPG and HER.
@article{kwon2024reinforcement,
title={Reinforcement Learning with Task Decomposition and Task-Specific Reward System for Automation of High-Level Tasks},
author={Kwon, Gunam and Kim, Byeongjun and Kwon, Nam Kyu},
journal={Biomimetics},
volume={9},
number={4},
pages={196},
year={2024},
publisher={MDPI}
}