エピソード

  • The Optimizer Becomes a Policy: Inside RP-1's Learned Planner for Robot World Models

    The Optimizer Becomes a Policy: Inside RP-1's Learned Planner for Robot World Models
    2026/09/03
    Pantheon Industries introduces Reinforced Planning (RP-1), the first fully learned planner that iteratively improves robot action plans from scratch using Reinforcement Learning over a frozen World Model. RP-1 reimagines planning as a learned policy: starting with a candidate action sequence, it uses a frozen World Model to imagine outcomes, a learned critic to evaluate closeness to goal (value-based scoring instead of latent distance), and a neural planner to revise the sequence over multiple iterations. This replaces hand-designed search heuristics (CEM, MPPI, Adam) with learned, reusable plan-improvement rules. Key results across ThreeRoom, OGBench, and Reacher benchmarks: RP-1 beats SOTA latent methods on 48 of 48 head-to-head comparisons, achieves 2.2x higher success rate on hard and long-horizon tasks, is 67x faster at scale (50 robot arms on H200 GPUs), and requires 1000x fewer World Model queries than MPPI/CEM.
    続きを読む 一部表示
    31 分
  • SAGA: Ground the Affordance, Then Learn the Motion

    SAGA: Ground the Affordance, Then Learn the Motion
    2026/09/03
    SAGA: Open-World Mobile Manipulation via Structured Affordance Grounding
    続きを読む 一部表示
    33 分
  • REMAC: Turning Robot Failures into Better Team Plans

    REMAC: Turning Robot Failures into Better Team Plans
    2026/09/03
    REMAC: Self-reflective and self-evolving multi-agent collaboration for long-horizon robot manipulation
    続きを読む 一部表示
    32 分
  • Restore the Sensor Contract: Training-Free Robustness for Brittle VLA Policies

    Restore the Sensor Contract: Training-Free Robustness for Brittle VLA Policies
    2026/09/02
    Vision–Language–Action (VLA) policies remain brittle under modest distribution shift. On LIBERO-Plus, contemporary models that solve clean tasks at high rates can fall below 30% success when the camera's viewpoint or background changes. This paper proposes a training-free method to improve out-of-distribution (OOD) robustness for VLA policies.
    続きを読む 一部表示
    31 分
  • Before the Robot Touches Anything: Physics-Grounded Deliberation with Embodied Tree of Thoughts

    Before the Robot Touches Anything: Physics-Grounded Deliberation with Embodied Tree of Thoughts
    2026/09/02
    Embodied Tree of Thoughts: Deliberate Manipulation Planning With Embodied World Model
    続きを読む 一部表示
    31 分
  • MemoAct: Why Robot Memory Needs Both a Scratchpad and an Archive

    MemoAct: Why Robot Memory Needs Both a Scratchpad and an Archive
    2026/09/01
    MemoAct: Atkinson–Shiffrin-Inspired Hierarchical Memory-Augmented Policy for Robotic Manipulation
    続きを読む 一部表示
    34 分
  • When Touch Means Something: OmniVTLA and the Missing Contact Layer in Robot Foundation Models

    When Touch Means Something: OmniVTLA and the Missing Contact Layer in Robot Foundation Models
    2026/09/01
    OmniVTLA: Vision-Tactile-Language-Action Models With Semantic-Aligned Tactile Sensing
    続きを読む 一部表示
    34 分
  • Gaussians as State, Futures as Supervision: Inside GSP-3D

    Gaussians as State, Futures as Supervision: Inside GSP-3D
    2026/08/31
    Introduction Learning robust visuomotor policies for bimanual manipulation remains challenging due to the stringent requirements for precise coordination between arms and the ability to generalize across diverse environmental conditions. Existing approaches struggle with generalization across diverse environmental conditions for long-horizon bimanual manipulation tasks.
    続きを読む 一部表示
    33 分