{
  "slug": "robottt-test-time-training-for-8k-timestep-robot-policy-context-715275",
  "title": "RoboTTT: Test-Time Training for 8K-Timestep Robot Policy Context",
  "dek": "RoboTTT uses test-time training and fast-weight recurrence to scale robot policy context to 8K timesteps without reported growth in inference latency.",
  "summary": "RoboTTT scales robot policy context to 8K timesteps with fast weights, reporting 87% real-robot gains and 62% over 1K context.",
  "tags": [
    "robotics",
    "robot-policies",
    "test-time-training",
    "long-context",
    "fast-weights",
    "vision-language-action",
    "manipulation",
    "closed-loop-control"
  ],
  "published_at": "2026-07-17T09:05:09.989+00:00",
  "grade": 9.1,
  "agent_utility": 9.4,
  "price_usdc": 0.157,
  "stats": {
    "claims": 16,
    "entities": 16,
    "faq": 10,
    "downloads": 0,
    "paid_downloads": 0
  },
  "preview": {
    "faq_questions": [
      "What is RoboTTT?",
      "How long is RoboTTT's context?",
      "Does RoboTTT increase inference latency?",
      "What are fast weights in RoboTTT?",
      "How is RoboTTT trained for long contexts?",
      "What capabilities does the abstract claim RoboTTT unlocks?",
      "How much better is RoboTTT than the reported baselines?",
      "What is the strongest long-horizon result in the abstract?",
      "Can RoboTTT be reproduced from the abstract alone?",
      "Why is the BibTeX field blank?"
    ],
    "entity_names": [
      {
        "name": "RoboTTT",
        "type": "model"
      },
      {
        "name": "Test-Time Training",
        "type": "technique"
      },
      {
        "name": "Fast weights",
        "type": "technique"
      },
      {
        "name": "Vision-Language-Action policies",
        "type": "model family"
      },
      {
        "name": "Sequence action forcing",
        "type": "training technique"
      },
      {
        "name": "Truncated backpropagation through time",
        "type": "training technique"
      },
      {
        "name": "8K-timestep context",
        "type": "metric"
      },
      {
        "name": "1K-timestep context",
        "type": "baseline condition"
      },
      {
        "name": "Single-step context baseline",
        "type": "baseline"
      },
      {
        "name": "State-of-the-art policies",
        "type": "baseline category"
      },
      {
        "name": "Closed-loop performance",
        "type": "evaluation metric"
      },
      {
        "name": "Human video demonstrations",
        "type": "demonstration source"
      },
      {
        "name": "Perturbations",
        "type": "evaluation condition"
      },
      {
        "name": "Real-robot manipulation tasks",
        "type": "benchmark setting"
      },
      {
        "name": "Five-minute, ten-stage assembly task",
        "type": "benchmark task"
      },
      {
        "name": "Project videos",
        "type": "resource"
      }
    ],
    "related_work_titles": [
      "Single-step and short-history robot foundation models",
      "Vision-Language-Action robot policies",
      "Test-Time Training for robot policies",
      "Recurrent sequence models for long-context control",
      "Long-sequence training with truncated backpropagation through time"
    ],
    "application_industries": [
      "robotics R&D",
      "real-robot manipulation research",
      "robot imitation learning research",
      "robot adaptation research",
      "robust robot control research"
    ],
    "glossary_terms": [
      "robot foundation model",
      "Vision-Language-Action policy",
      "visuomotor context",
      "robot policy",
      "Test-Time Training",
      "fast weights",
      "recurrent state",
      "gradient descent",
      "weight-space memory",
      "long-context conditioning",
      "sequence action forcing",
      "truncated backpropagation through time",
      "closed-loop performance",
      "one-shot in-context imitation",
      "human video demonstration",
      "on-the-fly policy improvement",
      "robustness to perturbations",
      "long-horizon task",
      "multi-stage task",
      "context length scaling"
    ]
  },
  "purchase": {
    "protocol": "x402",
    "version": 2,
    "network": "eip155:8453",
    "asset": "USDC",
    "price": 0.157,
    "full_markdown_url": "https://x402.disruptive-concepts.com/api/public/article/robottt-test-time-training-for-8k-timestep-robot-policy-context-715275/full",
    "full_json_url": "https://x402.disruptive-concepts.com/api/public/article/robottt-test-time-training-for-8k-timestep-robot-policy-context-715275/full?format=json"
  }
}