← Back ICRA 2026

An Alignment-Based Approach to Learning Motions from Demonstrations

Alex Cuellar, Christopher K Fourie, Julie A. Shah

PDF

AI summary

Key figure (auto-extracted from paper)

CALM enables robots to robustly learn and execute overlapping, multi-modal motions from few demonstrations by dynamically aligning with cluster-specific mean trajectories, overcoming the stability and overlap limitations of prior methods.

Learning from demonstration trajectory alignment movement primitives perturbation recovery multi-modal motion robotics

Problem

Existing learning-from-demonstration methods are either time-dependent, which causes undesirable trajectory snapping under perturbation, or time-independent, which cannot model overlapping trajectories, limiting robust real-world deployment.

Approach

CALM continuously maps a robot’s partial trajectory to a representative mean trajectory from clustered demonstrations, dynamically switching clusters when alignment shifts due to perturbations to ensure stable, accurate motion replication.

Key results

Proves global asymptotic stability and endpoint convergence for the alignment controller
Introduces an HMM-based alignment technique that handles perturbation-induced discontinuities
Successfully replicates overlapping and multi-modal trajectories on 2D datasets where baselines fail
Validates framework on a 7-DoF Franka robot across three physical domains

Why it matters

Provides a stable, overlap-capable learning framework for robots that can reliably recover from disturbances while executing complex, multi-modal tasks from minimal demonstrations.

Abstract

Learning from Demonstration (LfD) has shown to provide robots with fundamental motion skills for a variety of do- mains. Various branches of LfD research (e.g., learned dynamical systems and movement primitives) can generally be classified into “time-dependent” or “time-independent” systems. Each provides fundamental benefits and drawbacks – time-independent methods cannot learn overlapping trajectories, while time-dependence can result in undesirable behavior under perturbation. This letter introduces Cluster Alignment for Learned Motions (CALM), an LfD framework dependent upon an alignment with a representa- tive “mean” trajectory of demonstrated motions rather than pure time- or state-dependence. We discuss the convergence proper- ties of CALM, introduce an alignment technique able to handle the shifts in alignment possible under perturbation, and utilize demonstration clustering to generate multi-modal behavior. We show how CALM mitigates the drawbacks of time-dependent and time-independent techniques on 2D datasets and implement our system on a 7-DoF robot learning tasks in three domains.

Index terms

Learning from Demonstration Imitation Learning Probabilistic Inference