Failure-First Embodied AI

The authors propose Teach-and-Grow Learning (TGL), an agent-centered robotics architecture that extracts closed-loop skill blocks from demonstrations, stores execution repairs in experience memory, — In generalist robotics, Vision-Language-Action (VLA) models and World-Action Models (WAM) map visual observations and natural language instructions directly to physical low-level actions. While scaling neural network parameters and pretraining datasets has extended basic robot execution, s

The paper introduces MobileWorldSafety, a benchmark of 142 real Android application tasks designed to evaluate mobile GUI agent vulnerabilities to environmental injection attacks via a hybrid — Large Language Model (LLM)-powered Graphical User Interface (GUI) agents are rapidly evolving from conceptual research prototypes into autonomous assistants capable of operating mobile operating systems. By perceiving visual screen elements and executing cross-application interactions, these age

The authors present HAF, a framework combining hierarchical action-flow generation and spectral latent reinforcement learning to adapt pretrained generalist vision-language-action models to humanoid — Deploying pretrained generalist Vision-Language-Action (VLA) foundation models on high-dimensional humanoid robots introduces severe control challenges. While generalist VLA models have demonstrated strong multi-task capabilities on tabletop or fixed-base manipulators, applying them direc

The paper presents Robo-Dopamine 2.0, a process reward model that uses history conditioning and a signed progress space to distinguish valid progress, robustness variations, failures, and recovery — Large-scale Vision-Language-Action (VLA) models and imitation learning frameworks have advanced the capacity of robotic systems to execute manipulation tasks across diverse environments. However, long-horizon, contact-rich manipulation tasks remain highly susceptible to compounding executio

The authors propose RoboStriker, a hierarchical reinforcement learning framework that constrains multi-agent combat exploration to a learned latent motion manifold to prevent humanoid physical... — RoboStriker addresses the physical cold-start problem through a three-stage hierarchical architecture. The core design principle relies on decoupling high-level strategic reasoning from low-level motor execution. The pipeline first establishes a low-level physical foundation by learning to t

The paper investigates four mechanisms for coordinating dual joint-space and end-effector-space action heads in flow-matching policies to reduce false alarms in runtime kinematic failure detection,... — In modern robotic manipulation, generative action decoders—specifically diffusion models and conditional flow-matching (CFM) policies—have established themselves as premier frameworks for capturing complex, multimodal trajectory distributions. However, deploying these high-capacity poli

The authors present a framework that trains vision-language-action policies for urban navigation from automatically annotated egocentric web videos while characterizing and diagnosing long-tail... — Operating autonomous mobile agents in unconstrained urban environments presents severe safety challenges. Unlike indoor environments with static boundaries or structured roadways governed by lane markers and explicit traffic signals, open urban spaces are characterized by dense pedestrian i

The paper proposes a structure-guided spatiotemporal attention graph neural network (SGSAN) that soft-couples dynamic attention to a learned static directed dependency graph to improve traffic flow... — Deep spatiotemporal models integrating graph neural networks and dynamic attention mechanisms have demonstrated remarkable capabilities in modeling complex, non-linear traffic dynamics across network-scale infrastructure. However, deploying these unconstrained, black-box architectures i

The paper proposes an adaptive failure discovery method that uses control-variate residual modeling and a support-aware mutual-information objective to guide scenario selection on target systems... — Evaluating safety-critical autonomous systems—such as self-driving vehicles, robotic manipulators, and quadrupedal locomotion platforms—requires discovering rare, severe, and heterogeneous failure modes prior to deployment. However, identifying these edge cases through direct evaluation on

The authors propose ARMDIL, an ensemble framework that uses a multimodal large language model agent to dynamically route images to heterogeneous vision backbones across varied visual domains. — Modern image classification models achieve exceptional performance when trained and evaluated on isolated, task-specific datasets. However, real-world vision systems—such as those embedded in autonomous robotics, medical diagnostics, and mobile AI assistants—must operate across diverse visual en