← Back ICRA 2026

Task Generalization with Pathwise Conditioning of Gaussian Process for Learning from Demonstration

Adrian Prados, Gonzalo Espinoza, Alberto Mendez, Ramon Barber

PDF

AI summary

Key figure (auto-extracted from paper)

A zero-shot adaptation method using Gaussian Processes and Pathwise Conditioning enables real-time trajectory correction for robot learning from demonstration without retraining.

Gaussian Processes Learning from Demonstration Pathwise Conditioning Task Generalization Real-time Adaptation Movement Primitives

Problem

Learning from Demonstration algorithms struggle with out-of-distribution scenarios and covariant shifts, failing to generalize to new task parameters without costly retraining or extensive data augmentation.

Approach

The method learns movement primitives via Gaussian Processes with automatic kernel selection, then applies Pathwise Conditioning to instantly correct trajectories for new via-points in real-time without retraining.

Key results

Automatic kernel selection and heteroscedastic noise modeling for robust primitive learning
Real-time (30Hz) zero-shot adaptation to new via-points in position and orientation
Successful validation in simulations and on a physical robotic platform against state-of-the-art methods
Exact trajectory enforcement through new constraints while preserving prior uncertainty

Why it matters

Empowers robots to safely and reactively operate in dynamic human-centered environments by generalizing learned skills to unforeseen task variations without manual reprogramming.

Abstract

To effectively operate in human-centered environ- ments, robots must possess the capability to rapidly adapt to novel and changing situations. Techniques such as Learning from Demonstration enable fast learning without the need for explicit coding. However, in certain cases they exhibit limitations in generalizing beyond the set of demonstrations, which constrains their ability to rapidly adapt to unforeseen scenarios. In this work, we present a movement primitive learning algorithm based on Gaussian Processes, combined with a zero-shot adaptation to new via-points without requiring retraining, through Pathwise Conditioning. The algorithm not only learns the movement policy but is also capable of adapting it rapidly while preserving prior knowledge. The method has been evaluated through comparisons against other state-of- the-art approaches, experiments in simulated environments, as well as on a real robotic platform, generating new solutions for learned tasks by modifying via-points in both position and orientation. Website project: https://adrianprados. github.io/GaussianPathwiseLfD/.

Index terms

Learning from Demonstration Task and Motion Planning Imitation Learning