The field of robotic manipulation is witnessing significant advancements with the development of Vision-Language-Action (VLA) models. These models have demonstrated strong capabilities in learning complex behaviors from large-scale, multi-modal datasets, and have shown promising results in various robotic manipulation tasks. The current trend in VLA models is towards improving their robustness, generalizability, and efficiency. Researchers are exploring new architectures, training methods, and techniques to enhance the performance of VLA models in real-world scenarios. Notable papers in this area include SwiftVLA, which proposes a lightweight VLA model that achieves comparable performance to larger models while being 18 times faster and reducing memory footprint by 12 times. Another notable paper is VLASH, which introduces an asynchronous inference framework for VLA models that delivers smooth, accurate, and fast reaction control without additional overhead or architectural changes. Overall, the field of VLA models is rapidly evolving, and we can expect to see further innovations and improvements in the coming years. Noteworthy papers include SwiftVLA, which achieves state-of-the-art performance on edge devices, and VLASH, which enables fast and accurate reaction control for VLA models.
Vision-Language-Action Models for Robotic Manipulation
Sources
Sample-Efficient Expert Query Control in Active Imitation Learning via Conformal Prediction
Sigma: The Key for Vision-Language-Action Models toward Telepathic Alignment
SwiftVLA: Unlocking Spatiotemporal Dynamics for Lightweight VLA Models at Minimal Overhead
MM-ACT: Learn from Multimodal Parallel Generation to Act
CycleManip: Enabling Cyclic Task Manipulation via Effective Historical Perception and Understanding
VLASH: Real-Time VLAs via Future-State-Aware Asynchronous Inference
Real-World Reinforcement Learning of Active Perception Behaviors
LLM2Fx-Tools: Tool Calling For Music Post-Production
DiG-Flow: Discrepancy-Guided Flow Matching for Robust VLA Models
GR-RL: Going Dexterous and Precise for Long-Horizon Robotic Manipulation
Guardian: Detecting Robotic Planning and Execution Errors with Vision-Language Models
ManualVLA: A Unified VLA Model for Chain-of-Thought Manual Generation and Robotic Manipulation
SAM2Grasp: Resolve Multi-modal Grasping via Prompt-conditioned Temporal Action Prediction
Diagnose, Correct, and Learn from Manipulation Failures via Visual Symbols
Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach
VLA Models Are More Generalizable Than You Think: Revisiting Physical and Spatial Modeling
DAWZY: A New Addition to AI powered "Human in the Loop" Music Co-creation
GOMP: Grasped Object Manifold Projection for Multimodal Imitation Learning of Manipulation
Active Visual Perception: Opportunities and Challenges
PosA-VLA: Enhancing Action Generation via Pose-Conditioned Anchor Attention
Hierarchical Vision Language Action Model Using Success and Failure Demonstrations
Vision-Language-Action Models for Selective Robotic Disassembly: A Case Study on Critical Component Extraction from Desktops
MOVE: A Simple Motion-Based Data Collection Paradigm for Spatial Generalization in Robotic Manipulation
FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via neural Action Tokenization
STARE-VLA: Progressive Stage-Aware Reinforcement for Fine-Tuning Vision-Language-Action Models