Capek 0.5 is presented, an embodied vision-language model built around an execution-centric capability taxonomy that improves the large majority of matched benchmark rows over its initialization, retains all four specialized capabilities in one checkpoint with quantified losses, and transfers to closed-loop embodied task execution.
Abstract
Vision-language models are increasingly serving as the reasoning core of embodied agents. Robot execution is inherently iterative: each action reshapes the scene and physical state, continually renewing what must be perceived, reasoned about, and verified. Meeting these demands requires complementary capabilities that differ in supervision signals, prediction formats, and verification criteria. Existing approaches typically develop these capabilities against isolated, task-specific objectives, leaving open how they should be organized and integrated around execution as a whole. We present Capek 0.5, an embodied vision-language model built around an execution-centric capability taxonomy. Rather than organizing training by datasets or tasks, the taxonomy groups embodied capabilities according to their functional roles throughout execution and comprises four capability families: Spatial Reasoning, Temporal Understanding, Action Guidance, and State Verification. Each capability is first acquired by a dedicated specialist through reinforcement learning with verifiable rewards from a shared backbone, and the specialists are then consolidated into a single inference-time model through weight-space merging followed by routed policy-space distillation. We instantiate Capek 0.5 at the 2B and 35B-A3B scales and evaluate it from three complementary perspectives: comprehensive benchmark suites including Capek-StateBench, a new benchmark for state verification; a controlled study of capability retention from specialists to the unified model; and closed-loop evaluation in simulated embodied environments. Capek 0.5 improves the large majority of matched benchmark rows over its initialization, retains all four specialized capabilities in one checkpoint with quantified losses, and transfers to closed-loop embodied task execution.
This work presents PhyAgentOS, a runtime foundation delivering scheduling, verification, memory, benchmarking, and safety as system-level services, and distinguishes execution termination from semantic task completion via evidence-grounded verdicts of success, failure, or replan.
Yang Liu, Weixing Chen, Xinshuai Song et al.· 2 citations
Embodied artificial intelligence (Embodied AI) aims to develop agents that can perceive, reason, and act through continuous interaction with physical world. To achieve this goal, agents must not only understand visual scenes and language instructions, but also transform multimodal information into executable actions. This requirement calls for models that can integrate perception, semantic reasoning, and action generation within a unified decision‐making framework. Recent advances in large language models (LLMs) have provided important technical foundations for this integration, leading to emergence of vision‐language‐action (VLA) models as a central paradigm for embodied systems. Although existing surveys have reviewed Embodied AI from perspectives such as robotic systems, simulation environments and task settings, few have systematically analyzed action‐generation mechanisms of VLA models. This limitation makes it difficult to distinguish VLA architectures. To address this gap, we provide a comprehensive review of VLA models for Embodied AI from an action‐generation perspective. We first formulate VLA models as embodied decision‐making systems that map visual observations and language instructions to action sequences through interaction with dynamic environments. Building on this formulation, we review historical evolution of VLA models. Furthermore, we propose an action‐generation‐centered taxonomy that categorizes VLA models into three paradigms: direct policy learning, generative action modeling, and reasoning‐guided modeling. For each paradigm, we analyze representative methods, action‐generation mechanisms, advantages and limitations, thereby providing a coherent framework for comparing diverse VLA architectures. Beyond model architectures, we summarize key resources and benchmarking settings and discuss recent applications. Finally, we identify major challenges and future directions. By organizing VLA models around the central problem of action generation, we aim to provide a structured technical map for future research toward more general, reliable, and physically grounded Embodied AI.
Ning Xiong, Mingle Xu, Wei Chen et al.· Journal of Field Robotics· 0 citations
Embodied AI is moving from isolated perception or action modules toward physical agents that understand, plan under goals, act through robot bodies, monitor progress, and improve from experience. Existing systems address this loop only in parts: end-to-end policies generate actions but often lack spatial reasoning, planning, and execution assessment, while robot-agent systems orchestrate tools or specialists but do not learn a shared representation. This fragmentation limits general Physical Agentic AI. We present ACE-Brain-0.5, a unified embodied foundation model that organizes robot intelligence into five coupled functions: spatial perception, decision making, embodied interaction, self-monitoring, and self-improvement. Built on ACE-Brain-0, which established spatial intelligence as a shared scaffold across robot platforms, ACE-Brain-0.5 extends an understanding-centric model into a closed-loop foundation model. A single 8B backbone instantiates the first four functions: grounding objects and affordances, reasoning over 3D and egocentric spatial relations, decomposing instructions into subgoals, generating navigation and manipulation actions, and estimating progress for verification and recovery. To unify these capabilities without cross-task interference, we introduce SSR+, which extends Scaffold-Specialize-Reconcile with a Reactivate stage after task-vector merging. The fifth function, self-improvement, is realized by a companion framework that updates external execution state, including task schemas, spatial memory, and failure-recovery cases, from rollouts. Across fifteen benchmarks, ACE-Brain-0.5 improves over ACE-Brain-0 on 14 of 18 spatial perception and grounding benchmarks, achieves competitive navigation and manipulation performance, and provides strong progress estimation in ID and OOD settings. Together, these results mark an early step toward general Physical Agentic AI.
ACE-Brain Team Ziyang Gong, Haoming Gu, Zehang Luo et al.· 3 citations
Building capable embodied agents requires not only multimodal perception and understanding, but also agentic capabilities for reasoning about actions, adapting to evolving situations, and interacting with the physical world. In this report, we introduce Hy-Embodied-VLM-1.0, an efficient and powerful embodied foundation model specifically designed for embodied agents operating in the physical world. To cultivate such capabilities from the pre-training stage onward, we define an action-centric capability taxonomy comprising three progressive dimensions: Action-Relevant State Understanding, Action-Transition Reasoning, and Sequential and Adaptive Reasoning. Guided by this taxonomy, we develop a systematic data pipeline and curate data mixtures spanning both pre-training and post-training. To deliver strong physical-world understanding and interaction capabilities while supporting latency-sensitive deployment, we build our model on the Hy3-A3B language backbone and the Hy-ViT2 vision encoder. Its efficient Mixture-of-Experts architecture combines strong model capacity with high inference efficiency. We evaluate Hy-Embodied-VLM-1.0 on a comprehensive suite of 38 benchmarks covering embodied perception, physical-world understanding, and embodied reasoning. The model achieves the best performance among similarly sized models on 19 of the 38 benchmarks and substantially outperforms strong competitors, including Qwen3.6-A3B and Cosmos 3. Compared with the previous-generation Hy-Embodied-0.5 MoT-2B, Hy-Embodied-VLM-1.0 improves average performance by 8.4%. Despite activating only 3B parameters, it achieves performance close to that of the previous-generation model with 32B activated parameters. Beyond static benchmark evaluation, Hy-Embodied-VLM-1.0 also demonstrates strong performance on embodied agentic tasks requiring multi-turn interaction and long-horizon reasoning.
Ziyi Wang, Xumin Yu, Yongming Rao et al.· 4 citations
End-to-end vision-language-action (VLA) and world-action models offer an elegant route to general-purpose robotics, but their reliability is bounded by validated physical coverage. When an unfamiliar object, sensor, embodiment, or contact falls outside that coverage and no validated fallback exists, correcting the failure requires new robot data, a policy update, and regression testing. This recurring burden is the retraining tax. Unlike text, embodied data must often be created by operating machines. We present Teach-and-Grow Learning (TGL), an agent-centered architecture for general robot learning. In its general form, a multimodal agent turns a few successful demonstrations into reusable Skill Blocks: closed-loop behaviors for meaningful subgoals. In a new scene, the agent grounds and composes these blocks, selects learned or geometric tools, observes the physical outcome, and revises the route when execution departs from intent. A Skill Library stores executable behavior, while structured Experience Memory carries forward success, failure, and repair. New tasks are acquired without task-specific policy retraining. Our LIBERO evaluation attains state-of-the-art performance; controlled studies expose skill induction, persistent reuse, and agent-directed adaptation. Finally, we propose the Teach-and-Grow scaling-law hypothesis: if X denotes effective reusable experience, future-task error and teaching demand should approach irreducible floors as power laws in X. The architecture therefore treats deployment as a period of continued learning, in which one task can make the next easier.
Vision-and-Language Navigation (VLN) is a representative task in embodied artificial intelligence, requiring agents to perceive, understand, and make navigation decisions in partially observable environments according to natural language instructions. As research has expanded from early discrete simulation benchmarks to continuous control, interactive clarification, open-vocabulary perception, and real-world robotic deployment, VLN has evolved from a path-following multimodal task into an important research area connecting language understanding, environment modeling, spatial reasoning, and embodied execution. Existing surveys mainly organize the literature by timeline, model paradigm, or benchmark, while paying less attention to the internal components of VLN systems and their functional coupling. In this survey, we revisit VLN from a component-internal perspective, viewing it as a navigation system composed of internal components such as instructions, environment representations, and embodied agents, and organizing existing work around the functions and interactions of these components. Specifically, we summarize task definitions, datasets, and evaluation settings, and review representative methods and technical progress in instruction understanding and action generation, instruction–environment alignment, and robot–environment interaction understanding. We further discuss key trends as VLN moves from closed benchmarks toward open-world and real-world deployment, including reasoning-enhanced planning, open-vocabulary and online semantic mapping, long-horizon memory and structured spatial representation, and sim-to-real transfer across platforms. We hope this survey provides a clearer component-level analytical framework for understanding the evolution of internal VLN capabilities and for informing future method design and embodied-system deployment.