Search papers, labs, and topics across Lattice.
This paper introduces a novel structured demonstration collection strategy for Vision-Language-Action (VLA) learning, focusing on the organization of demonstrations to enhance policy learning efficiency and stability. By decomposing complex tasks into simpler sub-skills, standardizing environments, and progressively increasing task complexity, the authors show that this approach significantly improves the training outcomes for robotic manipulation tasks. Experimental results reveal that their method leads to higher success rates and greater stability compared to traditional end-to-end trajectory collection methods, underscoring the critical role of demonstration organization in effective imitation learning.
Organizing demonstrations into a simple-to-complex structure can dramatically boost the efficiency and stability of robotic manipulation learning.
Vision-Language-Action (VLA) models have demonstrated strong capabilities in robotic manipulation by integrating visual perception, language understanding, and robot action generation. Existing research has primarily focused on improving model architectures, training strategies, and dataset scale, while little attention has been paid to how demonstrations are collected and organized. We identify demonstration organization as a fundamental yet overlooked aspect of imitation learning, as it directly affects policy learning efficiency, training stability, and policy generalization. To address this gap, we propose a simple-to-complex structured demonstration collection strategy for VLA learning using a dual-arm robotic platform. Our approach systematically organizes data through three general principles: (i) decomposing complex manipulation tasks into progressively learnable sub-skills, (ii) standardizing the interaction environment to reduce unnecessary variability, and (iii) organizing demonstrations according to progressively increasing task complexity. This structured design enables VLA models to first acquire fundamental manipulation skills before learning increasingly complex task compositions, facilitating more effective learning of long-horizon manipulation tasks. We evaluate the proposed strategy on two representative robotic manipulation tasks: block grasping and sorting, and towel folding. Experimental results show consistent improvements in task success rate and training stability compared with the baseline method of directly collecting end-to-end complete task trajectories. These findings highlight demonstration organization as a previously underexplored but important factor in VLA learning and provide practical insights into efficient skill acquisition, scalable dataset construction, and long-horizon robotic manipulation.
