PG-IPRO：互動無障礙路線優化器

增強重點摘要

• •PG-IPRO utilizes a preference-based acquisition function that integrates human-in-the-loop feedback to dynamically prune the search space, specifically addressing the 'cold-start' problem in multi-objective pathfinding.
• •The algorithm leverages a constrained scalarization approach, allowing users to define 'soft' constraints on accessibility metrics (e.g., maximum incline or curb-cut density) that are updated in real-time during the interactive session.
• •Empirical evaluations demonstrate that PG-IPRO reduces computational overhead by approximately 40% compared to traditional evolutionary multi-objective optimization (EMO) algorithms by focusing search efforts only on user-preferred regions of the Pareto front.

競品分析

技術深入

• •Architecture: Employs a Gaussian Process (GP) surrogate model to approximate the underlying cost functions of urban accessibility metrics.
• •Acquisition Function: Uses a modified Expected Improvement (EI) variant that incorporates a preference vector derived from user interaction history.
• •Constraint Handling: Implements a penalty-based approach for non-accessible segments, where penalty weights are adjusted dynamically based on user-specified relaxation parameters.
• •Optimization Loop: Iterative refinement process where the surrogate model is updated after each user query, narrowing the search space to the local Pareto optimal region.

前景展望基於引用來源的 AI 分析

時間線