Tag
论文提出 Just Action Model (JAM)，将视觉-语言-动作模型与世界-动作模型解耦为骨干初始化、未来观测目标与推理机制三个可分离选择，并用仅在训练期使用的 foresight loss 在 LIBERO、RoboTwin 等基准上超越 FastWAM/ImageWAM，同时减少 73% 的训练 FLOPs。
The paper introduces Video-Point Model (VPM), a world model that jointly forecasts multi-view videos and 3D point trajectories from video histories and language instructions, improving 4D forecasting fidelity and robot manipulation success through shared 3D-aware representations.
GPT-6 Astra demonstrates advanced capabilities by controlling a humanoid robot in unfamiliar environments, remembering object locations, performing cleanup tasks, and retrieving items based on vague human requests using a Unitree G1 robot.
The paper introduces NowWAM, a future-target-free co-training formulation that denoises the current observation and predicts robot actions from the same visual stream, showing that a separate future visual target is not essential for adapting pretrained generative DiTs to robot control while achieving 87.7% on LIBERO-Plus with FLUX2-Klein and a 1.8x speedup.
LayerRoute introduces an action-conditioned routing interface for Vision-Language-Action policies that dynamically adapts access to VLM layer representations, improving robot manipulation performance with minimal additional parameters.
Show-Harness is a method that enables vision-language models to control robots through discrete semantic actions, allowing zero-shot deployment and efficient fine-tuning across different robots and GUIs.
MobileVLA-R1 2.0 is an RL-enhanced vision-language-action framework that couples structured reasoning with mobile robot control, achieving improvements in long-horizon instruction following on real-world platforms.
The paper introduces Safe Contrastive Reinforcement Learning (Safe-CRL) to address overestimation bias in goal-conditioned policy learning under failure termination, using one-bit failure signals to enhance safety and performance in robot navigation and locomotion tasks.
Perceptron Inc. has released Isaac 0.5, an open-source 36B parameter robotics foundation model that reduces teleoperation needs by 210X through action-free video scaling, enabling efficient fine-tuning for robot control.
LAWA is a world action model that uses latent actions to enable efficient future imagination for robot control, achieving state-of-the-art performance with reduced inference latency. It improves over baselines in generalization and efficiency without generating future observations.
Hydra-0 introduces action flow as a shared visual interface for generalist world modeling and robot control, achieving significant error reductions and enabling zero-shot composition across diverse embodiments.
French researchers connected a robot to brain cortex tissue donated from deceased individuals, demonstrating that isolated human brain tissue can still form new memories, learn new movements, and even learn to play new songs, potentially changing our understanding of brain death and viability.
GigaWorld-Policy-0.5 is an enhanced World Action Model for robot control that improves training and inference efficiency through a Mixed Action-Conditioned World Modeling strategy and a Mixture-of-Transformers architecture, achieving 85ms latency on a local RTX 4090.
LingBot-VA 2.0 is a video-action foundation model trained from scratch for robot control, achieving 225 Hz closed-loop execution with 13B parameters (1.9B active per token) and outperforming prior models on RoboTwin 2.0.
A robot using the LingBot-VA 2.0 video-action model picks objects off a moving conveyor belt in real-time at 1x speed, predicting future movements rather than reacting only to the current frame.
The article discusses the shift from reactive to prediction-based robot control, highlighted by the LingBot-VA 2.0 model which can keep up with fast-moving objects like an air-hockey puck and learn from few demonstrations.
LingBot-Video presents a DiT-based video pretraining framework with Mixture-of-Experts architecture, specialized data augmentation, and multi-dimensional reward system for embodied intelligence applications.
This paper identifies an embodiment gap in humanoid co-speech motion generation caused by human-centric pipelines, and proposes PhysDrift, an embodiment-aware framework that directly predicts executable humanoid joint trajectories from speech, improving speech-motion alignment and physical plausibility.
ImageWAM proposes replacing video generation with pretrained image editing models in world action models for robot control, achieving superior performance while reducing FLOPs to 1/6 and latency to 1/4 of video-based approaches.
LaWAM enables efficient robot control by predicting compact latent visual subgoals instead of expensive video generation, achieving state-of-the-art success rates with up to 24x lower latency than pixel-space world action models.