Embodied AI 101

Shaoqing Tan
  • 5,0 (1)
  • Technologies
  • Tous les jours

Stay in the loop on research in AI and physical intelligence.

  1. -6 h

    DART: One-Shot VLA Policy Adaptation via Weight-Space Arithmetic

    Uses weight-space arithmetic to isolate domain shifts from task knowledge, enabling one-shot VLA policy adaptation to new cameras or embodiments.

    23 min

  2. -6 h

    Does VLA Even Know the Basics? Act2Answer Benchmark

    Introduces a benchmark showing VLAs lose 20–40 points in commonsense/world knowledge versus source VLMs after robotics fine-tuning, evaluated via action-based answering.

    27 min

  3. -21 h

    Contact-Grounded Policy: Dexterous Visuotactile Policy with Generative Contact Grounding

    Learns dexterous manipulation policies that explicitly ground actions in generative contact predictions from visuotactile observations, improving robustness on contact-rich tasks.

    27 min

  4. -1 j

    Freeform Preference Learning (FPL) for Robotic Manipulation

    Introduces multi-axis preference supervision to learn dense, language-conditioned rewards across speed/precision/subtask axes without segmentation; enables compositional generalization and better long-horizon credit assignment than single-reward baselines.

    49 min

  5. -1 j

    Orca: The World is in Your Mind

    Proposes a general world foundation model leveraging Next-State-Prediction to jointly generate text, images, and embodied actions within a unified framework.

    29 min

  6. -1 j

    Qwen-RobotNav: A Scalable Unified Navigation Model for Agentic Robotics

    A unified 2B–8B parameter model for robot navigation tasks (VLN, ObjectNav, tracking, autonomous driving) via a configurable observation protocol, with demonstrated zero-shot deployment on real quadruped robots using agentic planners.

    37 min

  7. -5 j

    Scaling Robot Skills from Cheap Human Videos

    Replaces noisy 6-DoF hand poses with relative wrist translation as a shared action space between humans and bimanual robots, enabling scalable skill acquisition from inexpensive video data. This approach outperforms full-pose baselines for robot skill learning.

    15 min

  8. -6 j

    ABC: An Open Behavior Cloning Stack for Bimanual Manipulation

    Large-scale open-source framework for real-world robotic manipulation using behavior cloning, including the ABC-130K dataset with 3,500 hours and 130K+ episodes across 195 tasks. Provides hardware setups, simulators, and training recipes for Diffusion Transformers and Vision-Language-Action models.

    33 min
Tout afficher (169)

À propos

Stay in the loop on research in AI and physical intelligence.

Informations

  • Création
    Shaoqing Tan
  • Années d’activité
    2025 - 2026
  • Épisodes
    169
  • Classification
    Tous publics
  • Copyright
    © 2026 Shaoqing Tan

Vous aimeriez peut‑être aussi