Hyperparameter Optimization of Adaptive Simulated Annealing for Dynamic Path Planning in Sparse High-Dimensional Spaces

**Abstract:** This paper proposes a novel approach to hyperparameter optimization within Adaptive Simulated Annealing (ASA) algorithms for dynamic path planning in high-dimensional spaces characterized by sparse reward landscapes. Leveraging a meta-learning framework and Bayesian optimization, we autonomously tune ASA parameters – initial temperature, cooling schedule, move generation probability – to maximize the success rate and minimize path length in real-time evolving environments. Our solution addresses the critical challenge of maintaining robust performance with computationally efficient ASA configuration in environments where the optimal path is constantly shifting and the search space is both vast and sparsely populated. The methodology's scalability and adaptability are demonstrated through simulations within a randomized robotics navigation scenario, yielding a 1.7x improvement in average path success rate and a 12% reduction in average path length compared to standard hand-tuned ASA configurations.