Highly Configurable Evolutionary Multi‐Objective Optimisation and Application in UAV Path Planning

Traditional decomposition-based multi-objective evolutionary algorithms (such as MOEA/D) are highly effective for complex engineering problems like UAV path planning, but their performance heavily depends on manual parameter tuning and expert knowledge. To address these limitations, this research presents EMO-CRL, a highly configurable evolutionary multi-objective optimization framework driven by collaborative reinforcement learning. The framework establishes a comprehensive configuration space spanning multiple advanced MOEA/D mechanisms, which is systematically decomposed into six distinct sub-spaces to manage high-dimensional complexity. By formulating the dynamic algorithm configuration as a multi-agent Markov Decision Process (MDP), the system utilizes Value-Decomposition Networks (VDN) to enable collaborative learning and knowledge sharing among agents. The performance of EMO-CRL was validated on

standard WFG and DTLZ benchmarks, demonstrating superior optimization capabilities over state-of-the-art MOEA/D variants, and was successfully applied to solve multi-objective UAV path planning challenges. For M.Tech researchers, this framework provides a structured methodology to study adaptive guidance and navigation control. However, the source notes that high-dimensional configuration spaces require careful sub-space decomposition, and real-time computational overhead during online RL training remains a key factor for practical deployment.

Message: Hi FE, I need help with “Highly Configurable Evolutionary Multi‐Objective Optimisation and Application in UAV Path Planning” in “Aerospace Engineering”