DataPlatter: Boosting Robotic Manipulation Generalization with Minimal Costly Data

2025-03-25

AI Summary

This work addresses the limited generalization of Vision-Language-Action (VLA) models in robotic manipulation, particularly during the Spatial Reasoning Phase (SRP) over large workspaces, due to insufficient coverage of high-cost Physical Interaction Data (PIP). We propose a trajectory-phase decoupling and ratio-driven, subtask-specific training paradigm. Our approach innovatively decomposes manipulation trajectories into semantically meaningful, low-cost phases—such as SRP—that can be collected without physical interaction, and leverages their supervision signals to collaboratively enhance learning from scarce PIP. By integrating phased supervised learning with a ratio-based curriculum that prioritizes low-cost data guidance, our method significantly improves cross-object zero-shot transferability: it achieves up to a 41% absolute success rate gain in zero-shot settings and attains strong generalization with only minimal PIP.

Technology Category

• Intelligent Robots: Manipulation
• Computer Vision: Language and Vision
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

The growing adoption of Vision-Language-Action (VLA) models in embodied AI intensifies the demand for diverse manipulation demonstrations. However, high costs associated with data collection often result in insufficient data coverage across all scenarios, which limits the performance of the models. It is observed that the spatial reasoning phase (SRP) in large workspace dominates the failure cases. Fortunately, this data can be collected with low cost, underscoring the potential of leveraging inexpensive data to improve model performance. In this paper, we introduce the DataPlatter method, a framework that decouples training trajectories into distinct task stages and leverages abundant easily collectible SRP data to enhance VLA model's generalization. Through analysis we demonstrate that sub-task-specific training with additional SRP data with proper proportion can act as a performance catalyst for robot manipulation, maximizing the utilization of costly physical interaction phase (PIP) data. Experiments show that through introducing large proportion of cost-effective SRP trajectories into a limited set of PIP data, we can achieve a maximum improvement of 41% on success rate in zero-shot scenes, while with the ability to transfer manipulation skill to novel targets.

Problem

• Enhancing robotic manipulation generalization with low-cost data
• Addressing insufficient data coverage in Vision-Language-Action models
• Leveraging spatial reasoning phase data to boost model performance

Innovation

• Decouples training trajectories into task stages
• Leverages low-cost SRP data for model enhancement
• Maximizes utilization of costly PIP data