PROJECT CASE STUDY

Reinforcement Learning-Based Robot Navigation

AUTONOMOUS ROBOTICS · REINFORCEMENT LEARNING · ROS 2

Developed a reinforcement learning-based navigation framework for a simulated Clearpath Jackal J100 mobile robot using ROS 2 Humble and Gazebo Fortress. The project explores learning-based navigation across discrete and continuous environments, integrating LiDAR-based perception, robot state information and motion control. A Proximal Policy Optimization (PPO) policy was integrated with the simulated Jackal to generate continuous linear and angular velocity commands for navigation across different simulated environments.

View on GitHub ↗
Robot Navigation
OVERVIEW

Reinforcement Learning-Based Robot Navigation

Exploring reinforcement learning for robot navigation: learning how to move toward a goal through interaction with an environment.

01

Multiple Reinforcement Learning Algorithms

02

40-Dimensional LiDAR Observation Interface

03

ROS 2 and Gazebo integration

04

Multi-Environment Navigation

05

Motion control and safety

06

Classical and learning-based path planning