PhaseLoRA: Control-Regime-Conditioned Low-Rank Adaptation for Continuous-Action Vision-Language-Action Policies

2026-08-15

AI Summary

This study addresses the limitations of existing Vision-Language-Action (VLA) fine-tuning methods that overlook inter-phase dependencies in continuous manipulation and struggle with dynamic control via static adaptation. We propose PhaseLoRA, which integrates intra-trajectory phase conditioning into lightweight Parameter-Efficient Fine-Tuning (PEFT). By dynamically modulating LoRA’s left factor through weakly supervised descriptors, this method enables step-level adaptive and temporally conditioned action prediction. Evaluations on the LIBERO benchmark demonstrate that PhaseLoRA achieves a 12.2 percentage point increase in average success rate, significantly outperforming higher-rank LoRA variants with comparable parameter counts. These results validate the superiority of temporal awareness mechanisms in enhancing adaptability for continuous manipulation tasks, offering an effective solution for dynamic robotic control within resource-efficient fine-tuning frameworks.

Abstract

Parameter-efficient fine-tuning (PEFT) is a natural way to adapt pretrained vision-language-action (VLA) policies, but most adapter designs apply temporally static updates throughout a control rollout, overlooking the phase-dependent nature of continuous-action manipulation. Such policies traverse distinct regimes, including approach, contact transition, grasping, transport, and placement, each requiring different adaptation behaviors. We propose \textbf{PhaseLoRA}, a lightweight LoRA parameterization that conditions adaptation at each action-chunk prediction step using two weakly supervised descriptors: fine-control tendency and event/boundary intensity. PhaseLoRA modulates the LoRA left factor in the action expert, allowing the effective low-rank update direction to vary over time while keeping the backbone largely frozen. On LIBERO, PhaseLoRA improves average success rate by 12.2 points over a matched-parameter high-rank LoRA baseline and outperforms stronger LoRA variants. Ablations show that random temporal modulation and scalar gating do not reproduce the performance of the full model, while update-direction analyses reveal structured temporal variation associated with the predicted control descriptors. These results establish within-trajectory conditioning as an effective lightweight PEFT axis for continuous-action VLA policies.

Problem

• Vision-Language-Action Policies
• Parameter-Efficient Fine-Tuning
• Continuous-Action Manipulation
• Phase-Dependent Adaptation
• Low-Rank Adaptation

Innovation

• PhaseLoRA
• Vision-Language-Action
• Parameter-Efficient Fine-Tuning
• Temporal Modulation
• Continuous-Action Policies