Tag
The paper introduces Directed Temporal Representations for Control (DTRC), which learns a directed temporal quasimetric over frozen LeWorldModel features from offline visual trajectories to serve as a goal-relative progress signal for direct goal-conditioned policy learning, achieving strong control performance across ten visual tasks.
Reflex提出一个三阶段框架,让类人机器人通过自我中心RGB-D视觉实现对抛掷物体的全身心反射式接取控制,在仿真中达到85.7%成功率,并在真实Unitree G1机器人上以2.5米距离实现65%的接取成功率。
This paper introduces AnisoWM, a method that improves planning in JEPA world models by using anisotropic representations with learnable covariance, leading to better task-aligned planning costs and enhanced performance in visual control environments.
SenseNova-U1.5-8B-MoT is a native unified multimodal model for enhanced visual creation, featuring improvements in image generation quality, text rendering, and precise control.
Introduces Action-Conditioned Predictive Consistency (ACPC), a diagnostic for JEPA world models that measures how clean and perturbed observations diverge under action-conditioned rollouts, with theoretical bounds on prediction error and planner cost. Experiments on visual control tasks validate the diagnostic across models like LeWM and PLDM.
TaskSense introduces a task-centric world modeling framework that uses stochastic spatial attention conditioned on previous latent states and an auxiliary inverse-dynamics objective to focus on control-relevant regions, improving robustness to visual distractions compared to DreamerV3.
This paper proposes OG-SPR, a model-free visual RL algorithm that combines latent self-prediction with observation prediction to learn dynamics-aware representations, achieving improved sample efficiency on DeepMind Control Suite tasks.
Juggler is an AI coding agent that emphasizes visualization and controllability. It provides a native desktop app, session tree structure, plugin architecture, and supports multiple models like Claude, OpenAI, Gemini, etc., allowing developers to clearly view and edit every step of the LLM's actions.