# Optimus-3: Dual-Router Aligned Mixture-of-Experts Agent with Dual-Granularity Reasoning-Aware Policy Optimization

arXiv: https://arxiv.org/abs/2506.10357 · HTML: https://mlsanity.com/paper/2506.10357/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Placed in MoE in reinforcement learning and robotics, one of 12 directions in Mixture-of-experts.
- Builds on 33 of its 63 references that are in the map.
- First posted on arXiv 12 Jun 2025.

## Placed in

- Mixture-of-experts / [MoE in reinforcement learning and robotics](https://mlsanity.com/moe/moe-in-reinforcement-learning-and-robotics/)

## Builds on (most cited first, 30 of 33)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Continuous control with deep reinforcement learning](https://mlsanity.com/paper/1509.02971/) (2015)
- [Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://mlsanity.com/paper/2303.05499/) (2023)
- [Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context](https://mlsanity.com/paper/2403.05530/) (2024)
- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https://mlsanity.com/paper/2305.16291/) (2023)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [Mastering Diverse Domains through World Models](https://mlsanity.com/paper/2301.04104/) (2023)
- [DeepSeek-VL: Towards Real-World Vision-Language Understanding](https://mlsanity.com/paper/2403.05525/) (2024)
- [Object Hallucination in Image Captioning](https://mlsanity.com/paper/1809.02156/) (2018)
- [From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge](https://mlsanity.com/paper/2411.16594/) (2024)
- [Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents](https://mlsanity.com/paper/2302.01560/) (2023)
- [Hallucination of Multimodal Large Language Models: A Survey](https://mlsanity.com/paper/2404.18930/) (2024)
- [InternLM-XComposer2: Mastering Free-form Text-Image Composition and Comprehension in Vision-Language Large Model](https://mlsanity.com/paper/2401.16420/) (2024)
- [MineRL: A Large-Scale Dataset of Minecraft Demonstrations](https://mlsanity.com/paper/1907.13440/) (2019)
- [JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models](https://mlsanity.com/paper/2311.05997/) (2023)
- [Cradle: Empowering Foundation Agents Towards General Computer Control](https://mlsanity.com/paper/2403.03186/) (2024)
- [A Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, and Applications](https://mlsanity.com/paper/2503.07137/) (2025)
- [Video as the New Language for Real-World Decision Making](https://mlsanity.com/paper/2402.17139/) (2024)
- [Optimus-1: Hybrid Multimodal Memory Empowered Agents Excel in Long-Horizon Tasks](https://mlsanity.com/paper/2408.03615/) (2024)
- [Scaling Instructable Agents Across Many Simulated Worlds](https://mlsanity.com/paper/2404.10179/) (2024)
- [MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception](https://mlsanity.com/paper/2312.07472/) (2023)
- [MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts](https://mlsanity.com/paper/2407.21770/) (2024)
- [RL-GPT: Integrating Reinforcement Learning and Code-as-policy](https://mlsanity.com/paper/2402.19299/) (2024)
- [OmniJARVIS: Unified Vision-Language-Action Tokenization Enables Open-World Instruction Following Agents](https://mlsanity.com/paper/2407.00114/) (2024)
- [Odyssey: Empowering Minecraft Agents with Open-World Skills](https://mlsanity.com/paper/2407.15325/) (2024)

## Built on by (most cited first, 13 of 13)

- [A Survey of Reinforcement Learning for Large Reasoning Models](https://mlsanity.com/paper/2509.08827/) (2025)
- [Global Prior Meets Local Consistency: Dual-Memory Augmented Vision-Language-Action Model for Efficient Robotic Manipulation](https://mlsanity.com/paper/2602.20200/) (2026)
- [MAGNET: Towards Adaptive GUI Agents with Memory-Driven Knowledge Evolution](https://mlsanity.com/paper/2601.19199/) (2026)
- [ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation](https://mlsanity.com/paper/2605.05126/) (2026)
- [PhyAgentOS: A Self-Evolving Operating System for Embodied Agents with Decoupled Cognitive Planning and Physical Execution](https://mlsanity.com/paper/2607.16636/) (2026)
- [DETACH: Cross-domain Learning for Long-Horizon Tasks via Mixture of Disentangled Experts](https://mlsanity.com/paper/2508.07842/) (2025)
- [TINA: Text-Free Inversion Attack for Unlearned Text-to-Image Diffusion Models](https://mlsanity.com/paper/2603.17828/) (2026)
- [SPIKE: An Adaptive Dual Controller Framework for Cost-Efficient Long-Horizon Game Agents](https://mlsanity.com/paper/2605.18636/) (2026)
- [TINA+: Probing Residual Visual Knowledge in Unlearned Diffusion Models via Diffusion-Consistent Text-Free Inversion](https://mlsanity.com/paper/2608.17747/) (2026)
- [SurgGoal: Rethinking Surgical Planning Evaluation via Goal-Satisfiability](https://mlsanity.com/paper/2601.10455/) (2026)
- [Spark: Modular Spiking Neural Networks](https://mlsanity.com/paper/2602.02306/) (2026)
- [LookAgain: Closed-Loop GUI Grounding with Visually Grounded Reflection](https://mlsanity.com/paper/2608.09723/) (2026)
- [MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents](https://mlsanity.com/paper/2610.13829/) (2026)

## Abstract (arXiv)

Developing generalist agents capable of solving open-ended tasks in visually rich, dynamic environments remains a core pursuit of embodied AI. While Minecraft has emerged as a compelling benchmark, existing agents often suffer from fragmented cognitive abilities, lacking the synergy between reflexive execution (System 1) and deliberative reasoning (System 2). In this paper, we introduce Optimus-3, a generalist agent that organically integrates these dual capabilities within a unified framework. To achieve this, we address three fundamental challenges. First, to overcome the scarcity of reasoning data, we propose a Knowledge-Enhanced Automated Data Generation Pipeline. It synthesizes high-quality System 2 reasoning traces from raw System 1 interaction trajectories, effectively mitigating hallucinations via injection of domain knowledge. We release the resulting dataset, textbfOptimusM4, to the community. Second, to reconcile the dichotomous computational requirements of the dual systems, we design a Dual-Router Aligned MoE Architecture. It employs a Task Router to prevent task interference via parameter decoupling, and a Layer Router to dynamically modulate reasoning depth, creating a computational ``Fast Path'' for System 1 and a ``Deep Path'' for System 2. Third, to activate the reasoning capabilities of System 2, we propose Dual-Granularity Reasoning-Aware Policy Optimization (DGRPO) algorithm. It enforces Process-Outcome Co-Supervision via dual-granularity dense rewards, ensuring consistency between the thought process and the answer. Extensive evaluations demonstrate that Optimus-3 surpasses existing state-of-the-art methods on both System~2 (21% on Planning, 66% on Captioning, 76% on Embodied QA, 3.4× on Grounding, and 18% on Reflection) and System~1 (3% on Long-Horizon Action) tasks, with a notable 60% success rate on open-ended tasks.