Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Parallel Distributional Deep Reinforcement Learning for Mapless Navigation of Terrestrial Mobile Robots

Victor Augusto Kich, Alisson Henrique Kolling, Junior Costa de Jesus, Gabriel V. Heisler, Hiago Jacobs, Jair Augusto Bottega, André L. da S. Kelbouscas, Akihisa Ohya, Ricardo Bedin Grando, Paulo Lilles Jorge Drews-Jr, Daniel Fernando Tello Gamarra

TL;DR

Novel deep reinforcement learning techniques using parallel distributional actor-critic networks for navigating terrestrial mobile robots and results show that parallel distributional Deep-RL algorithms enhance decision-making and outperform non-distributional and behavior-based approaches in navigation and spatial generalization.

Abstract

This paper introduces novel deep reinforcement learning (Deep-RL) techniques using parallel distributional actor-critic networks for navigating terrestrial mobile robots. Our approaches use laser range findings, relative distance, and angle to the target to guide the robot. We trained agents in the Gazebo simulator and deployed them in real scenarios. Results show that parallel distributional Deep-RL algorithms enhance decision-making and outperform non-distributional and behavior-based approaches in navigation and spatial generalization.

Parallel Distributional Deep Reinforcement Learning for Mapless Navigation of Terrestrial Mobile Robots

TL;DR

Novel deep reinforcement learning techniques using parallel distributional actor-critic networks for navigating terrestrial mobile robots and results show that parallel distributional Deep-RL algorithms enhance decision-making and outperform non-distributional and behavior-based approaches in navigation and spatial generalization.

Abstract

This paper introduces novel deep reinforcement learning (Deep-RL) techniques using parallel distributional actor-critic networks for navigating terrestrial mobile robots. Our approaches use laser range findings, relative distance, and angle to the target to guide the robot. We trained agents in the Gazebo simulator and deployed them in real scenarios. Results show that parallel distributional Deep-RL algorithms enhance decision-making and outperform non-distributional and behavior-based approaches in navigation and spatial generalization.
Paper Structure (12 sections, 11 equations, 7 figures, 2 tables)

This paper contains 12 sections, 11 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Deep Reinforcement Learning
  5. Parallel Distributional Deterministic RL
  6. Parallel Distributional Stochastic RL
  7. Network Structure
  8. Reward Function
  9. Experiments
  10. Simulation and Real Scenarios
  11. Experimental Results
  12. Conclusion

Figures (7)

  • Figure 1: Left: Turtlebot3 Burger navigating a real-world obstacle scenario. Right: Input and output structure of our proposed PDDRL and PDSRL approaches.
  • Figure 2: Parallel Deep-RL training process structure.
  • Figure 3: Simulated and real setups.
  • Figure 4: Moving average of the agent's reward at each training step for all parallel approaches in each scenario. Scenarios organized by following the respectively order, from left to right and from top to bottom: 1, 2, 3, and 4.
  • Figure 5: The behavior of each parallel approach was evaluated by conducting 100 navigation trials in each simulated scenario. The lines illustrate the paths taken by the agents, with each agent given 25 attempts to capture each target.
  • ...and 2 more figures