Global Path Planning for Land–Air Amphibious Biomimetic Robot Based on Improved PPO

To address the path planning challenges for land–air amphibious biomimetic robots in unstructured environments, this study proposes a global path planning algorithm based on an Improved Proximal Policy Optimization (IPPO) framework. Unlike traditional single-domain navigation, amphibious robots face significant kinematic discontinuities when switching between terrestrial and aerial modes. To mitigate this, we integrate a Gated Recurrent Unit (GRU) module into the policy network, enabling the agent to capture temporal dependencies and make smoother decisions during mode transitions. Furthermore, to enhance exploration efficiency and stability, we replace the standard Gaussian noise with Ornstein–Uhlenbeck (OU) noise, which generates temporally correlated actions aligned with the robot’s physical inertia. Additionally, a Multi-Head Self-Attention mechanism is introduced to the value network, allowing the agent to dynamically prioritize critical environmental features—such as narrow obstacles—over irrelevant background noise. The simulation results demonstrate that the proposed IPPO algorithm significantly outperforms standard PPO baselines, achieving higher convergence speed, improved path smoothness, and greater success rates in complex amphibious scenarios.