# SafeDojo: Safe Reinforcement Learning for VLA via Interactive World Model

arXiv: https://arxiv.org/abs/2606.20698 · HTML: https://mlsanity.com/paper/2606.20698/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Latent dynamics models and model-based RL, one of 11 directions in World models.
- Builds on 18 of its 43 references that are in the map.
- First posted on arXiv 15 Jun 2026.

## Placed in

- World models / [Latent dynamics models and model-based RL](https://mlsanity.com/wm/latent-dynamics-models-and-model-based-rl/)
- Vision-language-action models / [Safety and adversarial attacks](https://mlsanity.com/vla/safety-and-adversarial-attacks/)

## Builds on (most cited first, 18 of 18)

- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [OpenVLA: An Open-Source Vision-Language-Action Model](https://mlsanity.com/paper/2406.09246/) (2024)
- [Wan: Open and Advanced Large-Scale Video Generative Models](https://mlsanity.com/paper/2503.20314/) (2025)
- [π0: A Vision-Language-Action Flow Model for General Robot Control](https://mlsanity.com/paper/2410.24164/) (2024)
- [Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success](https://mlsanity.com/paper/2502.19645/) (2025)
- [GRAPE: Generalizing Robot Policy via Preference Alignment](https://mlsanity.com/paper/2411.19309/) (2024)
- [SAFE: Multitask Failure Detection for Vision-Language-Action Models](https://mlsanity.com/paper/2506.09937/) (2025)
- [SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning](https://mlsanity.com/paper/2503.03480/) (2025)
- [WMPO: World Model-based Policy Optimization for Vision-Language-Action Models](https://mlsanity.com/paper/2511.09515/) (2025)
- [VLA-RFT: Vision-Language-Action Reinforcement Fine-tuning with Verified Rewards in World Simulators](https://mlsanity.com/paper/2510.00406/) (2025)
- [World-Env: Leveraging World Model as a Virtual Environment for VLA Post-Training](https://mlsanity.com/paper/2509.24948/) (2025)
- [SafeDreamer: Safe Reinforcement Learning with World Models](https://mlsanity.com/paper/2307.07176/) (2023)
- [WoVR: World Models as Reliable Simulators for Post-Training VLA Policies with RL](https://mlsanity.com/paper/2602.13977/) (2026)
- [World-VLA-Loop: Closed-Loop Learning of Video World Model and VLA Policy](https://mlsanity.com/paper/2602.06508/) (2026)
- [VLSA: Vision-Language-Action Models with Plug-and-Play Safety Constraint Layer](https://mlsanity.com/paper/2512.11891/) (2025)
- [Safe-Night VLA: Seeing the Unseen via Thermal-Perceptive Vision-Language-Action Models for Safety-Critical Manipulation](https://mlsanity.com/paper/2603.05754/) (2026)
- [FOSP: Fine-tuning Offline Safe Policy through World Models](https://mlsanity.com/paper/2407.04942/) (2024)
- [ThermoAct:Thermal-Aware Vision-Language-Action Models for Robotic Perception and Decision-Making](https://mlsanity.com/paper/2603.25044/) (2026)

## Built on by (most cited first, 2 of 2)

- [Coding Agents with an Obstacle-Aware Harness for Safe Robot Manipulation](https://mlsanity.com/paper/2609.20822/) (2026)
- [Towards Trustworthy Embodied Intelligence: A Systems Framework and Graded Trustworthiness Levels](https://mlsanity.com/paper/2607.26121/) (2026)

## Abstract (arXiv)

Safe control is a prerequisite for real-world embodied intelligence, for which safe reinforcement learning has emerged as a promising paradigm. However, existing safe reinforcement learning methods either require costly real-world exploration or depend on hand-crafted safety functions. Neither scales to vision-language-action models deployed in open-world physical environments. We propose SafeDojo, the first model-based safe reinforcement learning framework for vision-language-action policies designed to learn safe actions through world model-based imagination. Specifically, SafeDojo performs online reinforcement learning on top of an interactive video world model. The world model generates action-conditioned future predictions, from which a tailored ResNet success classifier estimates per-step task progress from imagined frames and a lightweight safety head predicts per-step safety costs from latent context together with the proposed action chunk, enabling simultaneous assessment of task execution and trajectory safety. The decoupled task-reward and safety-cost signals are balanced through a Lagrangian-based constrained GRPO objective, enabling coordinated improvement of task success and safety under explicit constraints. On SafeLIBERO, SafeDojo achieves the best aggregate task success, safe success, and execution efficiency among inference-time safety, model-free RL, and model-based RL baselines, with the best average safe-success rate on both levels and an 8.25 percentage-point improvement over the strongest baseline on Level I. Real-world Franka deployment further shows the best average task and safe-success rates across five tasks. Our results position world model-based safe reinforcement learning as a scalable and generalizable path toward safe embodied intelligence.