A Robust Task-Level Control Architecture for Learned Dynamical Systems

2025-11-12

AI Summary

In demonstration-based learning (LfD) for dynamical systems (DS), task-space trajectory tracking suffers from mismatches due to unmodeled dynamics, external disturbances, and system delays. Method: This paper proposes L1-DS, a task-level robust control architecture that synergistically integrates L1 adaptive control with a sliding-window dynamic time warping (DTW) target selection mechanism. The L1 adaptive module estimates and compensates for composite uncertainties in real time, while the DTW window dynamically aligns the current execution state with the reference trajectory’s phase. Contribution/Results: L1-DS ensures nominal stability while enabling online compensation for phase shifts and state deviations. Evaluated on the LASA and IROS handwriting datasets, it significantly improves trajectory tracking accuracy and phase consistency across diverse DS-LfD-generated motions. The approach establishes a new paradigm for robust operational-space LfD deployment.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Dynamical system (DS)-based learning from demonstration (LfD) is a powerful tool for generating motion plans in the operation (`task') space of robotic systems. However, the realization of the generated motion plans is often compromised by a''task-execution mismatch'', where unmodeled dynamics, persistent disturbances, and system latency cause the robot's actual task-space state to diverge from the desired motion trajectory. We propose a novel task-level robust control architecture, L1-augmented Dynamical Systems (L1-DS), that explicitly handles the task-execution mismatch in tracking a nominal motion plan generated by any DS-based LfD scheme. Our framework augments any DS-based LfD model with a nominal stabilizing controller and an L1 adaptive controller. Furthermore, we introduce a windowed Dynamic Time Warping (DTW)-based target selector, which enables the nominal stabilizing controller to handle temporal misalignment for improved phase-consistent tracking. We demonstrate the efficacy of our architecture on the LASA and IROS handwriting datasets.

Problem

• Addresses task-execution mismatch in robot motion plans
• Handles unmodeled dynamics and persistent disturbances in tracking
• Corrects temporal misalignment for phase-consistent trajectory tracking

Innovation

• L1 adaptive controller enhances DS-based LfD robustness
• Windowed DTW selector handles temporal misalignment in tracking
• Nominal stabilizing controller augments motion plan execution