Path planning plays a vital role in robotics and autonomous vehicles, enabling
efficient navigation toward a target. This project utilizes reinforcement learning
through trial and error to determine an optimal path. A custom environment will
be developed using OpenAI Gymnasium, where an agent navigates a scene. The
reward system encourages movement toward the target, incorporating checkpoints
to guide the agent in the desired direction.
The agent is trained using reinforcement learning algorithms, including Proximal
Policy Optimization (PPO), Deep Deterministic Policy Gradient (DDPG), Twin
Delayed DDPG (TD3), and Soft Actor-Critic (SAC). The results of these algo-
rithms are compared to evaluate their performance. Additionally, Optuna is used
for hyperparameter optimization, and the results are analyzed against manually
set hyperparameters. Finally, the trained models are tested, and the paths gener-
ated by the agents are examined.
The project successfully met its primary objectives, demonstrating the effective-
ness of the reinforcement learning algorithms in training agents to navigate the
environment. However, while the agents were able to generate feasible paths to-
ward their targets, there is still room for improvement in terms of path efficiency,
smoothness, and overall optimization. Future enhancements could focus on refin-
ing the reward function, incorporating more sophisticated exploration strategies,
or integrating additional constraints to improve the quality of the generated paths.
