Autonomous Navigation: Training and Deploying RL Agents in Simulated Environments
R. Pranay Kumar
Department of CSE (AI&ML)
2111cs020347@mallareddyuniversity.ac.in
P. Pranay
Department of CSE (AI&ML)
2111cs020349@mallareddyuniversity.ac.in
V. Pranay
Department of CSE (AI&ML)
2111cs020351@mallareddyuniversity.ac.in
K. Praneeth
Department of CSE (AI&ML)
2111cs020348@mallareddyuniversity.ac.in
T. Pranay
Department of CSE(AI&ML)
2111cs020350@mallareddyuniversity.ac.in
Prof. A. Vinnela
Department of CSE (AI&ML)
School of Engineering
MALLA REDDY UNIVERSITY
HYDERABAD
Abstract:-“This project focuses on training a reinforcement learning (RL) agent to autonomously navigate a simulated 3D environment using advanced RL algorithms such as Proximal Policy Optimization (PPO) or Q-Learning Algorithm . The environment is designed and simulated using Blender 3D, a versatile open-source 3D modeling and animation tool. The simulated environment includes dynamic obstacles, a target destination, and realistic physics, providing a challenging yet controlled setting for training the agent. The agent’s movements and interactions are controlled through Python scripting within Blender, enabling seamless integration of the RL framework with the 3D simulation. The agent learns to navigate the environment, avoid obstacles, and reach the target destination, guided by a reward-based system that incentivizes efficient and collision-free navigation. The training process involves defining the state space (e.g., agent position, target position, obstacle locations), action space (e.g., move forward, backward, left, right), and reward function (e.g., negative distance to target, penalties for collisions). The agent’s performance is evaluated based on metrics such as success rate, average time to reach the target, and robustness to dynamic obstacles. By leveraging Blender 3D and RL algorithms, this project demonstrates the effectiveness of simulation-based training for autonomous navigation tasks, providing a practical and scalable approach for exploring RL in dynamic and complex 3D environments. This work highlights the potential of combining 3D simulation and reinforcement learning for applications in robotics, game development, and AI research.”
Keywords: Reinforcement Learning, Autonomous Navigation, Proximal Policy Optimization (PPO), Q-Learning Algorithm, Blender 3D, 3D simulation, open-source software.
