{"slug": "what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and", "title": "What are the current mainstream and emerging policy architectures in robot imitation learning and VLA, and what are their limitations?", "summary": "A survey of robot imitation learning and vision-language-action (VLA) models finds that mainstream architectures — ACT-style imitation-trained transformers, Diffusion Policy, and VLAs built on a pretrained VLM backbone plus an action head (RT-2, OpenVLA, π0, π0.5, GR00T, SmolVLA) — remain limited by heavy demonstration requirements, poor performance under distribution shift, and vision-only policies that are blind to contact forces such as mass, friction and stiffness. The overview cites LIBERO benchmarks as near saturated at over 98%, saying they reveal little about real-world generalization, and points to emerging fixes including tactile and force-aware VLAs (Tactile-VLA, TacVLA, HapticVLA, the last of which distills force information so no force sensor is needed at inference), Real-Time Chunking for asynchronous action generation without retraining, and RL post-training on pretrained VLAs for reliability and data efficiency.", "body_md": "Welcome!  Short version, based on recent surveys and papers:\n\n**1. Mainstream**\n\n- Chunked policies trained by imitation: ACT-style transformers and Diffusion Policy.\n- VLAs: a pretrained VLM backbone plus an action head. Two families: autoregressive action tokens (RT-2, OpenVLA) and diffusion/flow-matching heads (π0, π0.5, GR00T, SmolVLA). Dual-system designs (a slow VLM planner plus a fast low-level policy) are common too. See this [survey](https://arxiv.org/abs/2509.19012) .\n\n**2. Limitations**\n\n- **Data:** VLAs need a lot of demos, and performance drops under distribution shift.\n- **Physical properties:** vision doesn’t show mass, friction or stiffness, so a vision-only VLA is[blind to contact forces](https://arxiv.org/pdf/2606.29089) .\n- **Latency:** big models are slow, so actions can be computed from stale observations.\n- **Evaluation:** benchmarks like LIBERO are near saturated (>98%), so they say little about real generalization ([survey](https://www.preprints.org/manuscript/202604.0664) ).\n\n**3. Emerging directions (matching your interests)**\n\n- **Tactile/force VLAs** for weight, friction and stiffness:[Tactile-VLA](https://arxiv.org/abs/2507.09160) ,[TacVLA](https://arxiv.org/abs/2603.12665) , and[HapticVLA](https://arxiv.org/abs/2603.15257) , which distills force info so no force sensor is needed at inference.\n- **Real-time execution:** Real-Time Chunking (asynchronous chunk generation, works with flow-based VLAs without retraining), and RL fine-tuning for real-time control ([Real-Time EXPO-FT](https://alphaxiv.org/abs/2609.18207) ).\n- **RL post-training** on top of pretrained VLAs for reliability and data efficiency.\n\nTip: you also posted this in Research, so keeping one thread will help you get answers in one place.", "url": "https://wpnews.pro/news/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and", "canonical_source": "https://discuss.huggingface.co/t/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-imitation-learning-and-vla-and-what-are-their-limitations/191715#post_2", "published_at": "2026-11-11 21:14:44+00:00", "updated_at": "2026-11-11 21:20:56.989909+00:00", "lang": "en", "topics": ["robotics", "machine-learning", "ai-research", "large-language-models", "ai-agents"], "entities": ["ACT", "Diffusion Policy", "RT-2", "OpenVLA", "π0", "GR00T", "SmolVLA", "LIBERO"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and", "markdown": "https://wpnews.pro/news/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and.md", "text": "https://wpnews.pro/news/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and.txt", "jsonld": "https://wpnews.pro/news/what-are-the-current-mainstream-and-emerging-policy-architectures-in-robot-and.jsonld"}}