# CureWM reduces optimism in robot world models

> A new fine‑tuning method cuts false‑success predictions on robot tasks from 80% to around 30% using counterfactual failures.

Oossa · 2026-10-08 · https://oossa.com/en/curewm-reduces-optimism-in-robot-world-models

Researchers introduced CureWM, a technique that adds simulated failure examples to existing robot world models. It builds alternative actions from successful demos, checks their outcomes in simulation or on a real arm, and fine‑tunes the model on these verified failures. On 484 held‑out LIBERO tasks, optimism dropped from 80% to 30‑43% across four fine‑tuned models. Physical robot tests showed false‑success scores fall from 90% to 33%.

## The facts

- Published: Thu Oct 08 2026 06:00:00 GMT+0200
- Optimism reduced to 30‑43% (mean 38%) after CureWM fine‑tuning

## Why it matters

Reduced false‑success predictions help robot planners avoid unsafe actions in real‑world deployments.

## Sources & references

1. [World Models Dream of Success: Diagnosing and Repairing Failure Insensitivity in Robot World Models](https://arxiv.org/abs/2610.09134) – arXiv cs.RO, 2026-10-08

Last updated: 2026-10-08