Topic
vision-language-action
New Training-Free Method Enables Robots to Follow Personalized Commands Like 'Bring My Cup'
Researchers propose Visual Attentive Prompting (VAP), a training-free perceptual adapter that enables vision-language-action models to follow personalized commands by using reference images as visual prompts. VAP outperforms generic policies and token-learning baselines on simulation and real-world benchmarks.
New Training-Free Method Compresses Vision-Language-Action Models by 50% Without Performance Loss
A research team led by Gia-Binh Ho et al. discovered that Vision-Language-Action (VLA) models exhibit severe layer-wise redundancy. They introduced a training-free compression pipeline using Centered Kernel Alignment to remove twin layers, achieving up to 50% depth reduction, 40-50% faster fine-tuning, and 30% faster inference while matching or exceeding full-scale performance.
New Benchmark and Method Address Occlusion in Vision-Language-Action Models for Robotics
Researchers introduced LIBERO-Occ, an occlusion-oriented benchmark for Vision-Language-Action (VLA) models, and proposed Viewpoint Imagination (VIM), a method that generates a complementary view from an occluded primary observation to condition action prediction. Experiments show that state-of-the-art VLAs suffer substantial performance degradation under occlusion, and VIM improves robustness across task suites, occlusion types, and severity levels without requiring additional cameras at deployment.
ResVLA Anchors Generative Policies with Residual Bridges to Reduce Noise and Speed Robot Learning
A team of researchers proposes ResVLA, a new architecture for generative Vision-Language-Action (VLA) policies that replaces the standard 'generation-from-noise' paradigm with a 'refinement-from-intent' approach. By using spectral analysis to separate robot motion into a deterministic low-frequency intent anchor and a stochastic high-frequency residual, the model achieves faster convergence, stronger robustness to perturbations, and competitive performance in both simulated and real-world robot experiments.
FineVLA Framework Improves Robot Instruction Following by 62.7% in Real-World Dual-Arm Manipulation
Researchers introduce FineVLA, an open framework for fine-grained instruction alignment in vision-language-action (VLA) robot policies. The framework includes a dataset of 47,159 human-verified trajectories, a benchmark with 500 videos and 11,631 atomic facts, and a steerable policy that improves real-world dual-arm manipulation success from 49.9% (raw-only) to 62.7%.
X-Tokenizer: Semantic Action Tokenizer Boosts Robot Control by 13.5% Over FAST
Researchers propose X-Tokenizer, a new action tokenizer that treats tokenization as semantic interface learning rather than mere compression. Using a lightweight encoder-Semantic Residual Quantization (SRQ)-decoder architecture, it improves multimodal grounding by 13.5% and long-horizon task performance by 8.25 points over existing methods like FAST.