Blog Tags GitHub

vla

18 posts tagged with "vla"

Next Forcing: Causal World Modeling with Multi-Chunk Prediction

多块预测（MCP）训练框架，通过轻量辅助模块同时去噪多个未来视频块，加速世界动作模型训练与推理

world-model video-generation vla diffusion-model

DIAL: Decoupling Intent and Action via Latent World Modeling

End-to-End VLA with Differentiable Latent Intent Bottleneck

vla robotics world-modeling latent-intent humanoid-robot manipulation vision-language-action

OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL

具有空间理解和在线强化学习的视觉-语言-动作模型

robotics diffusion-model reinforcement-learning vla vlm

FlashVLA: Think Twice, Act Once - Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models

首个免训练、即插即用的VLA推理加速框架，通过动作复用和视觉token剪枝实现55.7% FLOPs降低和36.0%延迟降低

efficient-attention pruning model-compression vla

EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models

一种无需训练的结构化推理加速框架，用于加速扩散式视觉-语言-动作模型

model-compression pruning llm-inference diffusion-model vla vlm

KV-Efficient VLA

A Method to Speed up Vision Language Models with RNN-Gated Chunked KV Cache

robotics kv-cache llm-inference vision-language vla vlm

Cosmos 3: Omnimodal World Models for Physical AI

NVIDIA全模态世界模型，统一语言、图像、视频、音频和动作的理解与生成，为Physical AI提供通用骨干网络

diffusion-model video-generation vision-language vla vlm world-model

OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models

全方向校准的LLM量化技术，通过LWC和LET实现PTQ效率与QAT性能的统一

quantization model-compression llm-inference vla

Qwen-VLA

统一的具身基础模型，通过DiT动作解码器将视觉-语言建模扩展到连续动作生成

robotics llm-training diffusion-model reinforcement-learning vla vlm

Cosmos-Predict2.5: World Simulation with Video Foundation Models for Physical

NVIDIA Cosmos世界基础模型2.5代：面向Physical AI的视频世界仿真

llm-training reinforcement-learning video-generation vla vlm

RL-VLA³: A Flexible and Asynchronous Reinforcement Learning Framework for VLA Training

面向VLA训练的灵活异步强化学习框架

reinforcement-learning system-optimization llm-training vla

Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Execution

小米机器人团队发布的先进VLA模型，通过精心设计的训练方案和部署策略实现高性能实时执行

robotics llm-training inference-serving diffusion-model vla vlm

KEEP: A KV-Cache-Centric Memory Management System for Efficient Embodied Planning

KV-Cache-centric memory management system for efficient embodied planning with static-dynamic memory construction, multi-hop recomputation, and layer-balanced loading

llm-inference kv-cache quantization vla prefix-caching

Fast-WAM: Do World Action Models Need Test-time Future Imagination?

世界动作模型是否需要测试时未来想象？

diffusion-model llm-training diffusion-transformer vla

KIVI: A Tuning-Free Asymmetric Quantization for KV Cache

面向KV缓存的免调优非对称量化

quantization llm-inference kv-cache vla

COAT: Compressing Optimizer States and Activation for Memory-Efficient FP8 Training

通过压缩优化器状态和激活实现高效FP8训练

quantization llm-training distributed-training vla vlm

November 22, 2025

MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots

通过强化学习增强视觉-语言-动作框架，实现四足机器人的显式推理和连续控制

llm-inference robotics reinforcement-learning vla

September 19, 2025

RLinf: Flexible and Efficient Large-scale Reinforcement Learning via Macro-to-Micro

强化学习（RL）在推进通用人工智能、具身智能和智能体智能方面展现出巨大潜力。然而，RL 工作流固有的异构性和动态性往往导致现有系统上硬件利用率低和训练缓慢。

system-optimization reinforcement-learning llm-training vla scheduling