Failure-First Embodied AI

Failure-First Embodied AI

Research audio from Failure-First: adversarial evaluation of embodied AI, jailbreak archaeology, policy analysis, and daily paper summaries from the AI safety frontier.

  1. 10小時前

    [Daily Paper] DocOps: A Verifiable Benchmark for Autonomous Agents in Complex Document Operations

    The authors introduce DocOps, a deterministically verifiable benchmark for evaluating autonomous agents on complex document operations, and analyze the operational failure modes of frontier models... — Large language model (LLM) agents are transitioning from "passive conversationalists" to active participants in digital workspaces, where they are increasingly tasked with manipulating complex artifacts such as spreadsheets, presentations, and PDFs. As Jiang et al. observe, while progres

  2. 1日前

    [Daily Paper] ReferTrack: Referring Then Tracking for Embodied Visual Tracking

    ReferTrack introduces a referring-then-tracking framework that grounds embodied visual tracking by selecting an explicit bounding box before decoding navigation waypoints from a single forward-facing... — Embodied Visual Tracking (EVT) requires a mobile agent to continuously pursue a specific target, defined by natural language instructions, using only egocentric onboard vision. Ye et al. characterize this task as a dual challenge of target identification—distinguishing the correct sub

  3. 2日前

    [Daily Paper] Clinical Pathways as Safety Specifications for Physical AI in Hospital Wards

    The authors propose a conceptual architecture that reinterprets clinical care pathways as runtime safety specifications for physical AI in hospital wards using constraint verification and multimodal... — The modern hospital ward represents one of the most complex deployment environments for Physical AI. Here, assistive robots and smart devices must operate in a high-stakes ecosystem of vulnerable patients, specialized medical staff, and rigorous clinical protocols. As these systems tra

  4. 3日前

    [Daily Paper] EgoRecovery: Acquiring Failure Recovery Ability Through Human Recovery Demonstration

    The authors introduce EgoRecovery, a framework that aligns scalable egocentric human recovery demonstrations to robot actions via a shared corrective-intent space and a learned recovery gate to... — In the field of imitation learning, standard training protocols often produce policies that excel only within a narrow "happy path"—a specific manifold of success demonstrated during data collection. However, these systems are notoriously fragile when deployed in unstructured real-world env

  5. 4日前

    [Daily Paper] The safety failures we are not instrumenting: a perspective on hidden safety-critical challenges in modern AI systems

    The authors propose a five-layer socio-technical framework—covering epistemic, control, temporal, organizational, and ecosystem integrity—to diagnose quiet, distributed, and under-instrumented... — Current discourse in AI safety frequently prioritizes visible, localized malfunctions—what Kasneci et al. (2026) characterize as the "tip of the iceberg." As illustrated in the authors' framework (Figure 1), these visible failures include shocking model outputs, explicit policy violations, a

  6. 9月4日

    [Daily Paper] StratMamba: Strategic and Reactive Stream Partitioning for Path-Efficient LiDAR-Based Obstacle Avoidance

    The authors propose StratMamba, a dual-stream state-space architecture that decouples fast-decay reactive LiDAR processing from slow-decay strategic goal tracking for quadruped robot navigation. — Autonomous quadrupedal navigation is defined by a fundamental temporal tension: the requirement for high-frequency, reactive obstacle avoidance balanced against the need for stable, long-horizon goal tracking. Modern controllers often struggle to reconcile these competing demands, frequently

  7. 9月3日

    [Daily Paper] MAMMOTH: A Multi-Modal End-to-End Policy for Off-Road Mobility Robust to Missing Modality

    The authors introduce MAMMOTH, an end-to-end off-road navigation policy that fuses multi-modal observations via a modality-dropout scheme and a diffusion-based trajectory planner to maintain safe... — Autonomous off-road navigation is currently hindered by extreme terrain diversity, drastic illumination changes, and the persistent threat of sensor degradation. Kotian et al. identify a significant "perception crisis" in the field, noting that while recent learning-based methods have mad

  8. 9月2日

    [Daily Paper] Directional Constraints for Efficient Exploration in Safe Reinforcement Learning

    The authors propose ATACOM-DC, an extension to the ATACOM safety layer that selectively enforces constraints in reinforcement learning by distinguishing between actions approaching versus moving away... — In the advancement of Safe Reinforcement Learning (SafeRL), a fundamental tension persists between providing strict safety guarantees and maintaining exploration efficiency. While safety violations in simulation are often inconsequential, deploying agents in real-world robotic environ

查看全部 (692)

關於

Research audio from Failure-First: adversarial evaluation of embodied AI, jailbreak archaeology, policy analysis, and daily paper summaries from the AI safety frontier.

資料

  • 創作者
    Failure-First Embodied AI
  • 活躍年份
    2026年
  • 單集
    692
  • 分級
    兒童適宜
  • 版權
    © 2026 Failure-First Embodied AI Research
  • 節目網站
    Failure-First Embodied AI