D squared-VLA: Dual-Memory Dual-Frequency Vision-Language-Action Model For Long Dynamic Manipulation
cs.CV
Submitted: 2026-09-28
Updated: 2026-09-30
Comments: 30 pages
License: http://creativecommons.org/licenses/by/4.0/
The gist: Long-horizon manipulation requires robots to remember cues that are no longer in view while responding to moving objects.
Terminology
Abstract
Long-horizon manipulation requires robots to remember cues that are no longer in view while responding to moving objects. Yet vision-language-action (VLA) policies often rely on the latest observation, and refreshing their visual context typically requires another costly vision-language model (VLM) pass. We present D squared-VLA, which combines dual memory and dual-frequency control at the KV-cache interface of a pretrained VLA. D squared-VLA uses block-wise causal KV caching to encode observations incrementally and, guided by distinct temporal attention patterns, constructs separate historical KV read views for the VLM and action expert. Between periodic VLM updates, a gated adapter incorporates fresh visual features into the latest history-conditioned KV block, while a short fast-memory queue supports action replanning. We introduce DOMINO-Long, a ten-task benchmark requiring robots to use earlier visual cues when manipulating moving objects. D squared-VLA achieves complete-task success rates of 29.3% on DOMINO, compared with 9.6% for π 0.5 and 17.2% for PUMA, and 60.0% on DOMINO-Long, compared with 35.4% and 20.6%, respectively. It improves success rates on eight real-robot tasks and reaches 97.5% on LIBERO-Long and 74.3% on RoboTwin 2.0.
Sources
- PaliGemma: A versatile 3B VLM for transfer
- Real-Time Execution of Action Chunking Flow Policies
- RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
- InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy
- Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
- StarVLA: A Lego-like Codebase for Vision-Language-Action Model Developing
- RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies
- SeedPolicy: Horizon Scaling via Self-Evolving Diffusion Policy for Robot Manipulation
- Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
- ReMem-VLA: Empowering Vision-Language-Action Model with Memory via Dual-Level Recurrent Queries
- CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulation
- Vision-Language Foundation Models as Effective Robot Imitators
- FAVLA: A Force-Adaptive Fast-Slow VLA model for Contact-Rich Robotic Manipulation
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation
- GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
- FAST: Efficient Action Tokenization for Vision-Language-Action Models
- SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
- MemER: Scaling Up Memory for Robot Control via Experience Retrieval
- TempoFit: Plug-and-Play Layer-Wise Temporal KV Memory for Long-Horizon Vision-Language-Action Manipulation
- VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models