N-Dimensional Reduction Algorithm for Learning from Demonstration Path Planning

This paper presents an n-dimensional reduction algorithm for Learning from Demonstration (LfD) for robotic path planning, addressing the complexity of highdimensional data. The method extends the Douglas–Peucker algorithm by incorporating velocity and orientation alongside position, enabling more pr...

Descripción completa

Detalles Bibliográficos
Autores: Manrique-Cordoba, Juliana, Casa-Lillo, Miguel Ángel de la, Sabater-Navarro, José María
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universidad Miguel Hernández de Elche
Repositorio:REDIUMH. Depósito Digital de la UMH
OAI Identifier:oai:dspace.umh.es:11000/38842
Acceso en línea:https://hdl.handle.net/11000/38842
Access Level:acceso abierto
Palabra clave:Learning from demonstration
Hidden Markov models
Data reduction
Douglas-Peucker algorithm
High-dimensional data encoding
Descripción
Sumario:This paper presents an n-dimensional reduction algorithm for Learning from Demonstration (LfD) for robotic path planning, addressing the complexity of highdimensional data. The method extends the Douglas–Peucker algorithm by incorporating velocity and orientation alongside position, enabling more precise trajectory simplification. A magnitude-based normalization process preserves proportional relationships across dimensions, and the reduced dataset is used to train Hidden Markov Models (HMMs), where continuous trajectories are discretized into identifier sequences. The algorithm is evaluated in 2D and 3D environments with datasets combining position and velocity. The results show that incorporating additional dimensions significantly enhances trajectory simplification while preserving key information. Additionally, the study highlights the importance of selecting appropriate encoding parameters to achieve optimal resolution. The HMM-based models generated new trajectories that retained the patterns of the original demonstrations, demonstrating the algorithm’s capacity to generalize learned behaviors for trajectory learning in high-dimensional spaces.