Continuous control of an underground loader using deep reinforcement\n learning

2021/03/01 by Sofi Backman, Backman, Sofi, Daniel Lindmark +9 · 3 citations
Engineering · #FOS: Computer and information sciences #Robotics (cs.RO) #Soft Robotics and Applications

paper · pdf · doi:10.48550/arxiv.2103.01283

openalex publication_date 2021/03/01 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28

Abstract

Reinforcement learning control of an underground loader is investigated in\nsimulated environment, using a multi-agent deep neural network approach. At the\nstart of each loading cycle, one agent selects the dig position from a depth\ncamera image of the pile of fragmented rock. A second agent is responsible for\ncontinuous control of the vehicle, with the goal of filling the bucket at the\nselected loading point, while avoiding collisions, getting stuck, or losing\nground traction. It relies on motion and force sensors, as well as on camera\nand lidar. Using a soft actor-critic algorithm the agents learn policies for\nefficient bucket filling over many subsequent loading cycles, with clear\nability to adapt to the changing environment. The best results, on average 75%\nof the max capacity, are obtained when including a penalty for energy usage in\nthe reward.\n

Cited by

• An integrated process for design and control of lunar robotics using AI and simulation
• Large Scale Robotic Material Handling: Learning, Planning, and Control
• A simulation framework for autonomous lunar construction work

Related

• Autonomous loading of ore piles with Load-Haul-Dump machines using Deep Reinforcement Learning
• Learning the Approach During the Short-loading Cycle Using Reinforcement Learning
• Learning to Capture Rocks using an Excavator: A Reinforcement Learning Approach with Guiding Reward Formulation
• 3D Operation of Autonomous Excavator based on Reinforcement Learning through Independent Reward for Individual Joints
• Towards continuous control of flippers for a multi-terrain robot using deep reinforcement learning