I am excited to share a glimpse of my final year thesis project Deep Reinforcement Learning (DQN) applied for autonomous navigation on TurtleBot3 in a simulated dynamic environment using Gazebo! This work focuses on training an agent to navigate efficiently by learning from its interactions with the environment no traditional mapping or path-planning needed!
𝐊𝐞𝐲 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬:
• Trained in dynamic environments with moving obstacles
• End-to-end learning approach without predefined maps
• Real-time reward feedback for obstacle avoidance and goal reaching
𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
• Designed and trained the agent using custom reward shaping
• Implemented exploration vs. exploitation balancing
• Achieved stable convergence and reliable path behavior