Robotics · 2026-10-01

RoboCoach: World Models as Active Coaches for Compositional Robot Skills

Jiajun Liu, Yifan Chen, Yichao Liu, Jiayi Zhang, Ruoqu Chen, Shaoxuan Xie, Guocai Yao, Mengdi Xu, Sen Cui, Changshui Zhang

arXiv:2609.39685 PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

RoboCoach is a world-model-guided coaching framework for compositional long-horizon robot skills. It uses a Route–Imagine–Diagnose–Improve (RIDI) loop: imagined rollouts in CoachWorld route control through reusable skill experts, a progress judge attributes the first unresolved subtask on timeouts, and a scorecard selects targeted subtask demonstrations and updates only the corresponding expert adapters.

Why it matters

The authors address the cost of adding end-to-end robot demonstrations by coupling world-model-imagined failures to selective supervision. Their framework decides both what to teach next (which subtask–expert pairs) and where to update (expert-specific LoRA adapters), aiming to improve complete-task success under limited demonstration budgets and to transfer to held-out compositions.

Method

RIDI loop: router selects next unfinished subtask and dispatches its reusable skill expert; Imagine rolls out the expert inside CoachWorld; Diagnose uses a progress judge to detect completion or record the first unresolved subtask–expert pair on timeout; Improve aggregates a scorecard to request targeted demonstrations and update the matched adapters.
Subtask–expert pair coaching: the router maps each subtask g to an expert e via η_b, making each atomic skill an explicit target for both data acquisition and adapter updates (instead of updating a single shared policy).
Priority via imagined failure records: the scorecard ranks subtask–expert pairs by task-balanced first-timeout mass and selects top pairs to request B_q/M demonstrations each, then performs LoRA fine-tuning for those experts using new data mixed with a replay sample.

Limitation

The authors note that trials without agreement among world-model seeds (at least two seeds agree) are excluded from the scorecard, meaning some imagined outcomes are not used for selecting coaching targets.

Abstract (from arXiv)

Long-horizon robot manipulation reuses skills across many task compositions, but improving these compositions with additional end-to-end demonstrations is costly. A practical self-improving system must decide both what to teach next and where to apply that supervision. We present ROBOCOACH, a world-model-guided coaching framework that uses imagined failures to guide demonstration requests and expert updates. Its Route-Imagine-Diagnose-Improve (RIDI) loop executes reusable skill experts inside COACHWORLD, our shared action-conditioned world model, and uses a progress judge to record the first subtask that fails to complete. Aggregated records select which subtask demonstrations to acquire and which expert adapters to update. Across two simulation suites and two real-robot platforms, imagined and deployed success correlate over 22 task-policy pairs (rho = 0.840). Controlled comparisons show that our coaching method outperforms matched baselines under matched data budgets and update schedules. 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. The coached experts also transfer to four held-out compositions, achieving an average success of 35.0%, compared with 0% for a shared-policy baseline updated with uniformly acquired demonstrations. Together, these results show that world models can serve as active coaches, turning imagined failures into targeted supervision for modular policy improvement. Project Page: https://robocoach-ai.github.io/

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-09-30 · Indexed by Pub-AI: 2026-10-01 UTC

preprint