RoboCoach: World Models as Active Coaches for Compositional Robot Skills

arXiv 2026 38.1 method, system, application

TLDR

A world-model-guided coaching framework uses imagined failures in a shared action-conditioned world model to target demonstrations and improve compositional robot skills.

Reasoning

The paper presents a clear framework with strong empirical support across simulations and real robots, including success correlations and controlled comparisons. Weaknesses are limited visibility into world model training details and potential limitations of the coaching approach from the abstract alone.

Read-first score

Read-first score 38.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,metric,result

Topical relevance 25%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=github

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 395.

Keyword Scores

world model

world dynamics prediction

world simulator

interactive world model

generative world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

ROBOCOACH: a world-model-guided coaching framework that uses imagined failures to decide what to teach next and where to apply supervision.
Route-Imagine-Diagnose-Improve (RIDI) loop: executes reusable skill experts inside COACHWORLD, a shared action-conditioned world model, and uses a progress judge to record the first subtask that fails to complete.
Aggregated imagined-failure records select which subtask demonstrations to acquire and which expert adapters to update.
Uses world models as active coaches for modular, compositional policy improvement, avoiding costly end-to-end demonstrations.

Methodology

ROBOCOACH uses a shared action-conditioned world model called COACHWORLD within a Route-Imagine-Diagnose-Improve loop: it routes imagined rollouts through reusable skill experts, diagnoses the first failing subtask with a progress judge, and aggregates failure records to select subtask demonstrations and expert adapter updates. The method is evaluated across two simulation suites and two real-robot platforms, with controlled comparisons against matched baselines under matched data budgets and update schedules. Metrics include imagined-versus-deployed success correlation and task success rates.

Key Results

Controlled comparisons show ROBOCOACH outperforms matched baselines under matched data budgets and update schedules, and imagined and deployed success correlate at rho = 0.840 over 22 task-policy pairs. With only 150 additional subtask demonstrations, success rises from 13.3% to 75.0% on Franka and from 40.0% to 83.8% on AgileX, while coached experts transfer to four held-out compositions with 35.0% average success versus 0% for a shared-policy baseline.

Resources

Paper