Robotic Grinding Skills Learning Based on Geodesic Length Dynamic Motion Primitives

Published 24 Apr 2025 in cs.RO | (2504.17216v1)

Abstract: Learning grinding skills from human craftsmen via imitation learning has become a key research topic in robotic machining. Due to their strong generalization and robustness to external disturbances, Dynamical Movement Primitives (DMPs) offer a promising approach for robotic grinding skill learning. However, directly applying DMPs to grinding tasks faces challenges, such as low orientation accuracy, unsynchronized position-orientation-force, and limited generalization for surface trajectories. To address these issues, this paper proposes a robotic grinding skill learning method based on geodesic length DMPs (Geo-DMPs). First, a normalized 2D weighted Gaussian kernel and intrinsic mean clustering algorithm are developed to extract geometric features from multiple demonstrations. Then, an orientation manifold distance metric removes the time dependency in traditional orientation DMPs, enabling accurate orientation learning via Geo-DMPs. A synchronization encoding framework is further proposed to jointly model position, orientation, and force using a geodesic length-based phase function. This framework enables robotic grinding actions to be generated between any two surface points. Experiments on robotic chamfer grinding and free-form surface grinding validate that the proposed method achieves high geometric accuracy and generalization in skill encoding and generation. To our knowledge, this is the first attempt to use DMPs for jointly learning and generating grinding skills in position, orientation, and force on model-free surfaces, offering a novel path for robotic grinding.

Abstract PDF Upgrade to Chat

Authors (6)

Summary

Robotic Grinding Skills Learning Based on Geo-DMPs

The paper "Robotic Grinding Skills Learning Based on Geodesic Length Dynamic Motion Primitives" proposes a novel approach to machine skill learning in robotic grinding tasks. Utilizing geodesic length Dynamic Motion Primitives (Geo-DMPs), the research addresses challenges in orientational accuracy, synchronization of trajectories, and the generalized applicability of motion tasks over free-form surfaces, moving beyond the conventional limitations of Dynamic Motion Primitives (DMPs).

Key Contributions and Theoretical Implications

Surface Encoding Methodology: The study introduces a two-dimensional weighted Gaussian kernel function combined with an intrinsic mean clustering algorithm for encoding the geometric features of free-form surfaces. This framework facilitates capturing spatial characteristics essential for robotic grinding, offering a systematic approach to trajectory planning without predefined surface designs.
Orientation Enhancement with Geo-DMPs: Geo-DMPs effectively remove the reliance on time factors typically seen in traditional DMPs, using geodesic distances on the orientation manifold for trajectory synchronization and learning. The introduction of geodesic length DMPs simplifies computational handling of quaternion data, mitigating issues related to time-dependence during trajectory shifts—a notable improvement in aligning orientation and positional trajectories over variable speeds.
Synchronization Framework: The paper elaborates on a synchronization strategy for position, orientation, and force through phase functions related to geodesic length. This synchronization paradigm provides an integrated approach to coordinate positional and orientational trajectories directly, enhancing the practical utility in precise robotic motion during grinding tasks.

Experimental Validation

Experiments conducted on chamfer grinding and free-form surface grinding tasks validate the framework. The proposed Geo-DMPs demonstrated reduced error margins in trajectory generation and orientation learning compared to existing Quat-DMPs and Riemannian-DMPs frameworks. The transition from demonstration-based skill acquisition to autonomous robotic execution further highlights the robustness of position-orientation-force synchronization.

Practical Implications

Robotic systems in manufacturing and processing applications demand precise handling of dynamic tasks over complex geometries, especially in sectors like aerospace and automotive. Geo-DMPs present an operational leap by encoding and executing grinding skills over free-form surfaces, achieving high geometric accuracy essential for varying industrial articulations. The reduction in setup time and manual intervention leads towards more agile, cost-effective robotic solutions—a crucial factor in customizing batch processing.

Vision for Future Research

The integration of real-time feedback loops within robotic controllers, potentially augmented by vision systems, is suggested in future developments. Exploring closed-loop control based on human-like sensory feedback could pave the way for adaptive, intelligent grinding mechanisms with real-time adjustments to processing outcomes, thereby closing the gap with human craftsmen capabilities.

This paper proposes a transformative avenue in robotic grinding skill acquisition, addressing inherent deficiencies in trajectory planning and execution while setting a foundation for broader applications in intelligent robotic systems.

Markdown Report Issue