embodied ai - 2025_06
Navigation
Home / Papers / embodied ai
- Part 1
Papers
The use of Multimodal Large Language Models (MLLMs) as an end-to-end solution for Embodied AI and Autonomous Driving has become a prevailing trend. While MLLMs have been extensively studied for visual semantic understanding tasks, their ability to perform precise and quantitative spatial-temporal understanding in real-world applications remains largely unexamined, leading to uncertain prospects. To evaluate models' Spatial-Temporal Intelligence, we introduce STI-Bench, a benchmark designed to evaluate MLLMs' spatial-temporal understanding through challenging tasks such as estimating and predicting the appearance, pose, displacement, and motion of objects. Our benchmark encompasses a wide range of robot and vehicle operations across desktop, indoor, and outdoor scenarios. The extensive experiments reveals that the state-of-the-art MLLMs still struggle in real-world spatial-temporal understanding, especially in tasks requiring precise distance estimation and motion analysis.
In dynamic 3D environments, accurately updating scene representations over time is crucial for applications in robotics, mixed reality, and embodied AI. As scenes evolve, efficient methods to incorporate changes are needed to maintain up-to-date, high-quality reconstructions without the computational overhead of re-optimizing the entire scene. This paper introduces CL-Splats, which incrementally updates Gaussian splatting-based 3D representations from sparse scene captures. CL-Splats integrates a robust change-detection module that segments updated and static components within the scene, enabling focused, local optimization that avoids unnecessary re-computation. Moreover, CL-Splats supports storing and recovering previous scene states, facilitating temporal segmentation and new scene-analysis applications. Our extensive experiments demonstrate that CL-Splats achieves efficient updates with improved reconstruction quality over the state-of-the-art. This establishes a robust foundation for future real-time adaptation in 3D scene reconstruction tasks.
Vision-Language Navigation (VLN) is a core challenge in embodied AI, requiring agents to navigate real-world environments using natural language instructions. Current language model-based navigation systems operate on discrete topological graphs, limiting path planning to predefined node connections. We propose VLN-R1, an end-to-end framework that leverages Large Vision-Language Models (LVLM) to directly translate egocentric video streams into continuous navigation actions, adopting GRPO-based training inspired by DeepSeek-R1. To enable effective training, we first construct the VLN-Ego dataset using a 3D simulator, Habitat, and propose Long-Short Memory Sampling to balance historical and current observations. While large language models can supervise complete textual instructions, they lack fine-grained action-level control. Our framework employs a two-stage training approach: a) Supervised fine-tuning (SFT) to align the model's action sequence text predictions with expert demonstrations, followed by b) Reinforcement fine-tuning (RFT) enhanced with a Time-Decayed Reward (TDR) mechanism that strategically weights multi-step future actions. Experimental results show VLN-R1 achieves strong performance on VLN-CE benchmark. VLN-R1 proves LVLMs can drive embodied navigation and enhance task-specific reasoning through data-efficient, reward-driven post-training.
World models have garnered increasing attention in the development of artificial general intelligence (AGI), serving as computational frameworks for learning representations of the external world and forecasting future states. While early efforts focused on 2D visual perception and simulation, recent 3D-aware generative world models have demonstrated the ability to synthesize geometrically consistent, interactive 3D environments, marking a shift toward 3D spatial cognition. Despite rapid progress, the field lacks systematic analysis to categorize emerging techniques and clarify their roles in advancing 3D cognitive world models. This survey addresses this need by introducing a conceptual framework, providing a structured and forward-looking review of world models transitioning from 2D perception to 3D cognition. Within this framework, we highlight two key technological drivers, particularly advances in 3D representations and the incorporation of world knowledge, as fundamental pillars. Building on these, we dissect three core cognitive capabilities that underpin 3D world modeling: 3D physical scene generation, 3D spatial reasoning, and 3D spatial interaction. We further examine the deployment of these capabilities in real-world applications, including embodied AI, autonomous driving, digital twin, and gaming/VR. Finally, we identify challenges across data, modeling, and deployment, and outline future directions for advancing more robust and generalizable 3D world models.
Multi-sensor fusion perception (MSFP) is a key technology for embodied AI, which can serve a variety of downstream tasks (e.g., 3D object detection and semantic segmentation) and application scenarios (e.g., autonomous driving and swarm robotics). Recently, impressive achievements on AI-based MSFP methods have been reviewed in relevant surveys. However, we observe that the existing surveys have some limitations after a rigorous and detailed investigation. For one thing, most surveys are oriented to a single task or research field, such as 3D object detection or autonomous driving. Therefore, researchers in other related tasks often find it difficult to benefit directly. For another, most surveys only introduce MSFP from a single perspective of multi-modal fusion, while lacking consideration of the diversity of MSFP methods, such as multi-view fusion and time-series fusion. To this end, in this paper, we hope to organize MSFP research from a task-agnostic perspective, where methods are reported from various technical views. Specifically, we first introduce the background of MSFP. Next, we review multi-modal and multi-agent fusion methods. A step further, time-series fusion methods are analyzed. In the era of LLM, we also investigate multimodal LLM fusion methods. Finally, we discuss open challenges and future directions for MSFP. We hope this survey can help researchers understand the important progress in MSFP and provide possible insights for future research.
Scientific discovery has long been constrained by human limitations in expertise, physical capability, and sleep cycles. The recent rise of AI scientists and automated laboratories has accelerated both the cognitive and operational aspects of research. However, key limitations persist: AI systems are often confined to virtual environments, while automated laboratories lack the flexibility and autonomy to adaptively test new hypotheses in the physical world. Recent advances in embodied AI, such as generalist robot foundation models, diffusion-based action policies, fine-grained manipulation learning, and sim-to-real transfer, highlight the promise of integrating cognitive and embodied intelligence. This convergence opens the door to closed-loop systems that support iterative, autonomous experimentation and the possibility of serendipitous discovery. In this position paper, we propose the paradigm of Intelligent Science Laboratories (ISLs): a multi-layered, closed-loop framework that deeply integrates cognitive and embodied intelligence. ISLs unify foundation models for scientific reasoning, agent-based workflow orchestration, and embodied agents for robust physical experimentation. We argue that such systems are essential for overcoming the current limitations of scientific discovery and for realizing the full transformative potential of AI-driven science.
Editing illumination in long videos with complex dynamics has significant value in various downstream tasks, including visual content creation and manipulation, as well as data scaling up for embodied AI through sim2real and real2real transfer. Nevertheless, existing video relighting techniques are predominantly limited to portrait videos or fall into the bottleneck of temporal consistency and computation efficiency. In this paper, we propose TC-Light, a novel paradigm characterized by the proposed two-stage post optimization mechanism. Starting from the video preliminarily relighted by an inflated video relighting model, it optimizes appearance embedding in the first stage to align global illumination. Then it optimizes the proposed canonical video representation, i.e., Unique Video Tensor (UVT), to align fine-grained texture and lighting in the second stage. To comprehensively evaluate performance, we also establish a long and highly dynamic video benchmark. Extensive experiments show that our method enables physically plausible relighting results with superior temporal coherence and low computation cost. The code and video demos are available at https://dekuliutesla.github.io/tclight/.
Recent benchmarks and datasets have been proposed to improve spatial reasoning in vision-language models (VLMs), yet existing open resources remain limited in scale, visual diversity, and instruction expressiveness. In this work, we introduce InternSpatial, the largest open-source dataset for spatial reasoning in VLMs, along with InternSpatial-Bench, a corresponding evaluation benchmark designed to assess spatial understanding under diverse instruction formats. InternSpatial comprises 12 million QA pairs spanning both single-view and multi-view settings, drawn from diverse visual environments and supporting 19 instruction formats that reflect varied query styles. For evaluation, we propose InternSpatial-Bench for single-view tasks and expand multi-view reasoning by introducing a novel rotation angle prediction task that has not been explored in prior work. Experimental results show that models trained on InternSpatial achieve 12.1% improvement on InternSpatial-Bench and 10.7% on VSI-Bench, while maintaining strong performance on general-purpose benchmarks. We hope these resources will support the development of spatially capable VLMs in practical applications such as robotics and embodied AI.
Telerobotics is a key foundation in autonomous Industrial Cyber-Physical Systems (ICPS), enabling remote operations across various domains. However, conventional cloud-based telerobotics suffers from latency, reliability, scalability, and resilience issues, hindering real-time performance in critical applications. Cloud-Fog Telerobotics (CFTel) builds on the Cloud-Fog Automation (CFA) paradigm to address these limitations by leveraging a distributed Cloud-Edge-Robotics computing architecture, enabling deterministic connectivity, deterministic connected intelligence, and deterministic networked computing. This paper synthesizes recent advancements in CFTel, aiming to highlight its role in facilitating scalable, low-latency, autonomous, and AI-driven telerobotics. We analyze architectural frameworks and technologies that enable them, including 5G Ultra-Reliable Low-Latency Communication, Edge Intelligence, Embodied AI, and Digital Twins. The study demonstrates that CFTel has the potential to enhance real-time control, scalability, and autonomy while supporting service-oriented solutions. We also discuss practical challenges, including latency constraints, cybersecurity risks, interoperability issues, and standardization efforts. This work serves as a foundational reference for researchers, stakeholders, and industry practitioners in future telerobotics research.
World models -- generative models that simulate environment dynamics conditioned on past observations and actions -- are gaining prominence in planning, simulation, and embodied AI. However, evaluating their rollouts remains a fundamental challenge, requiring fine-grained, temporally grounded assessment of action alignment and semantic consistency -- capabilities not captured by existing metrics. Vision-Language Models (VLMs) have shown promise as automatic evaluators of generative content due to their strong multimodal reasoning abilities. Yet, their use in fine-grained, temporally sensitive evaluation tasks remains limited and requires targeted adaptation. We introduce a evaluation protocol targeting two recognition tasks -- action recognition and character recognition -- each assessed across binary, multiple-choice, and open-ended formats. To support this, we present UNIVERSE (UNIfied Vision-language Evaluator for Rollouts in Simulated Environments), a method for adapting VLMs to rollout evaluation under data and compute constraints. We conduct a large-scale study comparing full, partial, and parameter-efficient finetuning across task formats, context lengths, sampling strategies, and data compositions. The resulting unified evaluator matches the performance of task-specific baselines using a single checkpoint. Human studies confirm strong alignment with human judgments, establishing UNIVERSE as a scalable, semantics-aware evaluator for world models.
Embodied artificial intelligence (Embodied AI) plays a pivotal role in the application of advanced technologies in the intelligent era, where AI systems are integrated with physical bodies that enable them to perceive, reason, and interact with their environments. Through the use of sensors for input and actuators for action, these systems can learn and adapt based on real-world feedback, allowing them to perform tasks effectively in dynamic and unpredictable environments. As techniques such as deep learning (DL), reinforcement learning (RL), and large language models (LLMs) mature, embodied AI has become a leading field in both academia and industry, with applications spanning robotics, healthcare, transportation, and manufacturing. However, most research has focused on single-agent systems that often assume static, closed environments, whereas real-world embodied AI must navigate far more complex scenarios. In such settings, agents must not only interact with their surroundings but also collaborate with other agents, necessitating sophisticated mechanisms for adaptation, real-time learning, and collaborative problem-solving. Despite increasing interest in multi-agent systems, existing research remains narrow in scope, often relying on simplified models that fail to capture the full complexity of dynamic, open environments for multi-agent embodied AI. Moreover, no comprehensive survey has systematically reviewed the advancements in this area. As embodied AI rapidly evolves, it is crucial to deepen our understanding of multi-agent embodied AI to address the challenges presented by real-world applications. To fill this gap and foster further development in the field, this paper reviews the current state of research, analyzes key contributions, and identifies challenges and future directions, providing insights to guide innovation and progress in this field.
Vision-Language Navigation (VLN) is a core challenge in embodied AI, requiring agents to navigate real-world environments using natural language instructions. Current language model-based navigation systems operate on discrete topological graphs, limiting path planning to predefined node connections. We propose VLN-R1, an end-to-end framework that leverages Large Vision-Language Models (LVLM) to directly translate egocentric video streams into continuous navigation actions, adopting GRPO-based training inspired by DeepSeek-R1. To enable effective training, we first construct the VLN-Ego dataset using a 3D simulator, Habitat, and propose Long-Short Memory Sampling to balance historical and current observations. While large language models can supervise complete textual instructions, they lack fine-grained action-level control. Our framework employs a two-stage training approach: a) Supervised fine-tuning (SFT) to align the model's action sequence text predictions with expert demonstrations, followed by b) Reinforcement fine-tuning (RFT) enhanced with a Time-Decayed Reward (TDR) mechanism that strategically weights multi-step future actions. Experimental results show VLN-R1 achieves strong performance on VLN-CE benchmark. VLN-R1 proves LVLMs can drive embodied navigation and enhance task-specific reasoning through data-efficient, reward-driven post-training.
Flawed planning from VLM-driven embodied agents poses significant safety hazards, hindering their deployment in real-world household tasks. However, existing static, non-interactive evaluation paradigms fail to adequately assess risks within these interactive environments, since they cannot simulate dynamic risks that emerge from an agent's actions and rely on unreliable post-hoc evaluations that ignore unsafe intermediate steps. To bridge this critical gap, we propose evaluating an agent's interactive safety: its ability to perceive emergent risks and execute mitigation steps in the correct procedural order. We thus present IS-Bench, the first multi-modal benchmark designed for interactive safety, featuring 161 challenging scenarios with 388 unique safety risks instantiated in a high-fidelity simulator. Crucially, it facilitates a novel process-oriented evaluation that verifies whether risk mitigation actions are performed before/after specific risk-prone steps. Extensive experiments on leading VLMs, including the GPT-4o and Gemini-2.5 series, reveal that current agents lack interactive safety awareness, and that while safety-aware Chain-of-Thought can improve performance, it often compromises task completion. By highlighting these critical limitations, IS-Bench provides a foundation for developing safer and more reliable embodied AI systems.
Developing embodied agents capable of performing complex interactive tasks in real-world scenarios remains a fundamental challenge in embodied AI. Although recent advances in simulation platforms have greatly enhanced task diversity to train embodied Vision Language Models (VLMs), most platforms rely on simplified robot morphologies and bypass the stochastic nature of low-level execution, which limits their transferability to real-world robots. To address these issues, we present a physics-based simulation platform DualTHOR for complex dual-arm humanoid robots, built upon an extended version of AI2-THOR. Our simulator includes real-world robot assets, a task suite for dual-arm collaboration, and inverse kinematics solvers for humanoid robots. We also introduce a contingency mechanism that incorporates potential failures through physics-based low-level execution, bridging the gap to real-world scenarios. Our simulator enables a more comprehensive evaluation of the robustness and generalization of VLMs in household environments. Extensive evaluations reveal that current VLMs struggle with dual-arm coordination and exhibit limited robustness in realistic environments with contingencies, highlighting the importance of using our simulator to develop more capable VLMs for embodied tasks. The code is available at https://github.com/ds199895/DualTHOR.git.
Open-vocabulary 3D object detection has gained significant interest due to its critical applications in autonomous driving and embodied AI. Existing detection methods, whether offline or online, typically rely on dense point cloud reconstruction, which imposes substantial computational overhead and memory constraints, hindering real-time deployment in downstream tasks. To address this, we propose a novel reconstruction-free online framework tailored for memory-efficient and real-time 3D detection. Specifically, given streaming posed RGB-D video input, we leverage Cubify Anything as a pre-trained visual foundation model (VFM) for single-view 3D object detection by bounding boxes, coupled with CLIP to capture open-vocabulary semantics of detected objects. To fuse all detected bounding boxes across different views into a unified one, we employ an association module for correspondences of multi-views and an optimization module to fuse the 3D bounding boxes of the same instance predicted in multi-views. The association module utilizes 3D Non-Maximum Suppression (NMS) and a box correspondence matching module, while the optimization module uses an IoU-guided efficient random optimization technique based on particle filtering to enforce multi-view consistency of the 3D bounding boxes while minimizing computational complexity. Extensive experiments on ScanNetV2 and CA-1M datasets demonstrate that our method achieves state-of-the-art performance among online methods. Benefiting from this novel reconstruction-free paradigm for 3D object detection, our method exhibits great generalization abilities in various scenarios, enabling real-time perception even in environments exceeding 1000 square meters.
Visual Navigation is a core task in Embodied AI, enabling agents to navigate complex environments toward given objectives. Across diverse settings within Navigation tasks, many necessitate the modelling of sequential data accumulated from preceding time steps. While existing methods perform well, they typically process all historical observations simultaneously, overlooking the internal association structure within the data, which may limit the potential for further improvements in task performance. We address this by examining the unique characteristics of Navigation tasks through the lens of causality, introducing a causal framework to highlight the limitations of conventional sequential methods. Leveraging this insight, we propose Causality-Aware Navigation (CAN), which incorporates a Causal Understanding Module to enhance the agent's environmental understanding capability. Empirical evaluations show that our approach consistently outperforms baselines across various tasks and simulation environments. Extensive ablations studies attribute these gains to the Causal Understanding Module, which generalizes effectively in both Reinforcement and Supervised Learning settings without computational overhead.
Robots are increasingly envisioned as human companions, assisting with everyday tasks that often involve manipulating deformable objects. Although recent advances in robotic hardware and embodied AI have expanded their capabilities, current systems still struggle with handling thin, flat, and deformable objects such as paper and fabric. This limitation arises from the lack of suitable perception techniques for robust state estimation under diverse object appearances, as well as the absence of planning techniques for generating appropriate grasp motions. To bridge these gaps, this paper introduces PP-Tac, a robotic system for picking up paper-like objects. PP-Tac features a multi-fingered robotic hand with high-resolution omnidirectional tactile sensors \sensorname. This hardware configuration enables real-time slip detection and online frictional force control that mitigates such slips. Furthermore, grasp motion generation is achieved through a trajectory synthesis pipeline, which first constructs a dataset of finger's pinching motions. Based on this dataset, a diffusion-based policy is trained to control the hand-arm robotic system. Experiments demonstrate that PP-Tac can effectively grasp paper-like objects of varying material, thickness, and stiffness, achieving an overall success rate of 87.5\%. To our knowledge, this work is the first attempt to grasp paper-like deformable objects using a tactile dexterous hand. Our project webpage can be found at: https://peilin-666.github.io/projects/PP-Tac/
Visual Language Navigation (VLN) is a fundamental task within the field of Embodied AI, focusing on the ability of agents to navigate complex environments based on natural language instructions. Despite the progress made by existing methods, these methods often present some common challenges. First, they rely on pre-trained backbone models for visual perception, which struggle with the dynamic viewpoints in VLN scenarios. Second, the performance is limited when using pre-trained LLMs or VLMs without fine-tuning, due to the absence of VLN domain knowledge. Third, while fine-tuning LLMs and VLMs can improve results, their computational costs are higher than those without fine-tuning. To address these limitations, we propose Weakly-supervised Partial Contrastive Learning (WPCL), a method that enhances an agent's ability to identify objects from dynamic viewpoints in VLN scenarios by effectively integrating pre-trained VLM knowledge into the perception process, without requiring VLM fine-tuning. Our method enhances the agent's ability to interpret and respond to environmental cues while ensuring computational efficiency. Experimental results have shown that our method outperforms the baseline methods on multiple benchmarks, which validate the effectiveness, robustness and generalizability of our method.
Intelligent embodied agents (e.g. robots) need to perform complex semantic tasks in unfamiliar environments. Among many skills that the agents need to possess, building and maintaining a semantic map of the environment is most crucial in long-horizon tasks. A semantic map captures information about the environment in a structured way, allowing the agent to reference it for advanced reasoning throughout the task. While existing surveys in embodied AI focus on general advancements or specific tasks like navigation and manipulation, this paper provides a comprehensive review of semantic map-building approaches in embodied AI, specifically for indoor navigation. We categorize these approaches based on their structural representation (spatial grids, topological graphs, dense point-clouds or hybrid maps) and the type of information they encode (implicit features or explicit environmental data). We also explore the strengths and limitations of the map building techniques, highlight current challenges, and propose future research directions. We identify that the field is moving towards developing open-vocabulary, queryable, task-agnostic map representations, while high memory demands and computational inefficiency still remaining to be open challenges. This survey aims to guide current and future researchers in advancing semantic mapping techniques for embodied AI systems.
Despite recent advances in multimodal content generation enabled by vision-language models (VLMs), their ability to reason about and generate structured 3D scenes remains largely underexplored. This limitation constrains their utility in spatially grounded tasks such as embodied AI, immersive simulations, and interactive 3D applications. We introduce a new paradigm that enables VLMs to generate, understand, and edit complex 3D environments by injecting a continually evolving spatial context. Constructed from multimodal input, this context consists of three components: a scene portrait that provides a high-level semantic blueprint, a semantically labeled point cloud capturing object-level geometry, and a scene hypergraph that encodes rich spatial relationships, including unary, binary, and higher-order constraints. Together, these components provide the VLM with a structured, geometry-aware working memory that integrates its inherent multimodal reasoning capabilities with structured 3D understanding for effective spatial reasoning. Building on this foundation, we develop an agentic 3D scene generation pipeline in which the VLM iteratively reads from and updates the spatial context. The pipeline features high-quality asset generation with geometric restoration, environment setup with automatic verification, and ergonomic adjustment guided by the scene hypergraph. Experiments show that our framework can handle diverse and challenging inputs, achieving a level of generalization not observed in prior work. Further results demonstrate that injecting spatial context enables VLMs to perform downstream tasks such as interactive scene editing and path planning, suggesting strong potential for spatially intelligent systems in computer graphics, 3D vision, and embodied applications. Project page: https://spatctxvlm.github.io/project_page/.
Constructing a physically realistic and accurately scaled simulated 3D world is crucial for the training and evaluation of embodied intelligence tasks. The diversity, realism, low cost accessibility and affordability of 3D data assets are critical for achieving generalization and scalability in embodied AI. However, most current embodied intelligence tasks still rely heavily on traditional 3D computer graphics assets manually created and annotated, which suffer from high production costs and limited realism. These limitations significantly hinder the scalability of data driven approaches. We present EmbodiedGen, a foundational platform for interactive 3D world generation. It enables the scalable generation of high-quality, controllable and photorealistic 3D assets with accurate physical properties and real-world scale in the Unified Robotics Description Format (URDF) at low cost. These assets can be directly imported into various physics simulation engines for fine-grained physical control, supporting downstream tasks in training and evaluation. EmbodiedGen is an easy-to-use, full-featured toolkit composed of six key modules: Image-to-3D, Text-to-3D, Texture Generation, Articulated Object Generation, Scene Generation and Layout Generation. EmbodiedGen generates diverse and interactive 3D worlds composed of generative 3D assets, leveraging generative AI to address the challenges of generalization and evaluation to the needs of embodied intelligence related research. Code is available at https://horizonrobotics.github.io/robot_lab/embodied_gen/index.html.
Leveraging vast amounts of unlabeled internet video data for embodied AI is currently bottlenecked by the lack of action labels and the presence of action-correlated visual distractors. Although recent latent action policy optimization (LAPO) has shown promise in inferring proxy-action labels from visual observations, its performance degrades significantly when distractors are present. To address this limitation, we propose a novel object-centric latent action learning framework that centers on objects rather than pixels. We leverage self-supervised object-centric pretraining to disentangle action-related and distracting dynamics. This allows LAPO to focus on task-relevant interactions, resulting in more robust proxy-action labels, enabling better imitation learning and efficient adaptation of the agent with just a few action-labeled trajectories. We evaluated our method in eight visually complex tasks across the Distracting Control Suite (DCS) and Distracting MetaWorld (DMW). Our results show that object-centric pretraining mitigates the negative effects of distractors by 50%, as measured by downstream task performance: average return (DCS) and success rate (DMW).
Recently, latent action learning, pioneered by Latent Action Policies (LAPO), have shown remarkable pre-training efficiency on observation-only data, offering potential for leveraging vast amounts of video available on the web for embodied AI. However, prior work has focused on distractor-free data, where changes between observations are primarily explained by ground-truth actions. Unfortunately, real-world videos contain action-correlated distractors that may hinder latent action learning. Using Distracting Control Suite (DCS) we empirically investigate the effect of distractors on latent action learning and demonstrate that LAPO struggle in such scenario. We propose LAOM, a simple LAPO modification that improves the quality of latent actions by 8x, as measured by linear probing. Importantly, we show that providing supervision with ground-truth actions, as few as 2.5% of the full dataset, during latent action learning improves downstream performance by 4.2x on average. Our findings suggest that integrating supervision during Latent Action Models (LAM) training is critical in the presence of distractors, challenging the conventional pipeline of first learning LAM and only then decoding from latent to ground-truth actions.
Constructing a physically realistic and accurately scaled simulated 3D world is crucial for the training and evaluation of embodied intelligence tasks. The diversity, realism, low cost accessibility and affordability of 3D data assets are critical for achieving generalization and scalability in embodied AI. However, most current embodied intelligence tasks still rely heavily on traditional 3D computer graphics assets manually created and annotated, which suffer from high production costs and limited realism. These limitations significantly hinder the scalability of data driven approaches. We present EmbodiedGen, a foundational platform for interactive 3D world generation. It enables the scalable generation of high-quality, controllable and photorealistic 3D assets with accurate physical properties and real-world scale in the Unified Robotics Description Format (URDF) at low cost. These assets can be directly imported into various physics simulation engines for fine-grained physical control, supporting downstream tasks in training and evaluation. EmbodiedGen is an easy-to-use, full-featured toolkit composed of six key modules: Image-to-3D, Text-to-3D, Texture Generation, Articulated Object Generation, Scene Generation and Layout Generation. EmbodiedGen generates diverse and interactive 3D worlds composed of generative 3D assets, leveraging generative AI to address the challenges of generalization and evaluation to the needs of embodied intelligence related research. Code is available at https://horizonrobotics.github.io/robot_lab/embodied_gen/index.html.
Embodied navigation stands as a foundation pillar within the broader pursuit of embodied AI. However, previous navigation research is divided into different tasks/capabilities, e.g., ObjNav, ImgNav and VLN, where they differ in task objectives and modalities, making datasets and methods are designed individually. In this work, we take steps toward generalist navigation agents, which can follow free-form instructions that include arbitrary compounds of multi-modal and multi-capability. To achieve this, we propose a large-scale benchmark and corresponding method, termed OctoNav-Bench and OctoNav-R1. Specifically, OctoNav-Bench features continuous environments and is constructed via a designed annotation pipeline. We thoroughly craft instruction-trajectory pairs, where instructions are diverse in free-form with arbitrary modality and capability. Also, we construct a Think-Before-Action (TBA-CoT) dataset within OctoNav-Bench to provide the thinking process behind actions. For OctoNav-R1, we build it upon MLLMs and adapt it to a VLA-type model, which can produce low-level actions solely based on 2D visual observations. Moreover, we design a Hybrid Training Paradigm (HTP) that consists of three stages, i.e., Action-/TBA-SFT, Nav-GPRO, and Online RL stages. Each stage contains specifically designed learning policies and rewards. Importantly, for TBA-SFT and Nav-GRPO designs, we are inspired by the OpenAI-o1 and DeepSeek-R1, which show impressive reasoning ability via thinking-before-answer. Thus, we aim to investigate how to achieve thinking-before-action in the embodied navigation field, to improve model's reasoning ability toward generalists. Specifically, we propose TBA-SFT to utilize the TBA-CoT dataset to fine-tune the model as a cold-start phrase and then leverage Nav-GPRO to improve its thinking ability. Finally, OctoNav-R1 shows superior performance compared with previous methods.
Embodied foundation models are crucial for Artificial Intelligence (AI) interacting with the physical world by integrating multi-modal inputs, such as proprioception, vision and language, to understand human intentions and generate actions to control robots. While these models demonstrate strong generalization and few-shot learning capabilities, they face significant challenges in continually acquiring new skills without forgetting previously learned skills, a problem known as catastrophic forgetting. To address this issue, we propose the Analytic Task Scheduler (ATS), a novel framework for continual learning in embodied foundation models. ATS consists of a task-specific model library, where each model is fine-tuned independently on a single task, and an analytic scheduler trained using recursive least squares (RLS) to learn the mapping between language instructions and task-specific models. This architecture enables accurate task recognition and dynamic model selection while fundamentally avoiding parameter interference across tasks. The scheduler updates its parameters incrementally using only statistics (autocorrelation and cross-correlation matrices), enabling forgetting-resistant learning without the need to revisit historical data. We validate ATS on a real-world robot platform (RM65B), demonstrating superior resistance to forgetting and strong adaptability to task variations. The results highlight ATS as an effective, scalable, and deployable solution for continual learning in embodied foundation models operating in complex, dynamic environments. Our code will be available at https://github.com/MIAA-Embodied-AI/AnalyticTaskScheduler
Large Vision-Language Action (VLA) models have shown significant potential for embodied AI. However, their predominant training via supervised fine-tuning (SFT) limits generalization due to susceptibility to compounding errors under distribution shifts. Reinforcement learning (RL) offers a path to overcome these limitations by optimizing for task objectives via trial-and-error, yet a systematic understanding of its specific generalization benefits for VLAs compared to SFT is lacking. To address this, our study introduces a comprehensive benchmark for evaluating VLA generalization and systematically investigates the impact of RL fine-tuning across diverse visual, semantic, and execution dimensions. Our extensive experiments reveal that RL fine-tuning, particularly with PPO, significantly enhances generalization in semantic understanding and execution robustness over SFT, while maintaining comparable visual robustness. We identify PPO as a more effective RL algorithm for VLAs than LLM-derived methods like DPO and GRPO. We also develop a simple recipe for efficient PPO training on VLAs, and demonstrate its practical utility for improving VLA generalization. The project page is at https://rlvla.github.io
Vision-and-Language Navigation (VLN) presents a complex challenge in embodied AI, requiring agents to interpret natural language instructions and navigate through visually rich, unfamiliar environments. Recent advances in large vision-language models (LVLMs), such as CLIP and Flamingo, have significantly improved multimodal understanding but introduced new challenges related to computational cost and real-time deployment. In this project, we propose a modular, plug-and-play navigation framework that decouples vision-language understanding from action planning. By integrating a frozen vision-language model, Qwen2.5-VL-7B-Instruct, with lightweight planning logic, we aim to achieve flexible, fast, and adaptable navigation without extensive model fine-tuning. Our framework leverages prompt engineering, structured history management, and a two-frame visual input strategy to enhance decision-making continuity across navigation steps. We evaluate our system on the Room-to-Room benchmark within the VLN-CE setting using the Matterport3D dataset and Habitat-Lab simulation environment. Although our initial results reveal challenges in generalizing to unseen environments under strict evaluation settings, our modular approach lays a foundation for scalable and efficient navigation systems, highlighting promising directions for future improvement through enhanced environmental priors and expanded multimodal input integration.
Coordinating multiple embodied agents in dynamic environments remains a core challenge in artificial intelligence, requiring both perception-driven reasoning and scalable cooperation strategies. While recent works have leveraged large language models (LLMs) for multi-agent planning, a few have begun to explore vision-language models (VLMs) for visual reasoning. However, these VLM-based approaches remain limited in their support for diverse embodiment types. In this work, we introduce VIKI-Bench, the first hierarchical benchmark tailored for embodied multi-agent cooperation, featuring three structured levels: agent activation, task planning, and trajectory perception. VIKI-Bench includes diverse robot embodiments, multi-view visual observations, and structured supervision signals to evaluate reasoning grounded in visual inputs. To demonstrate the utility of VIKI-Bench, we propose VIKI-R, a two-stage framework that fine-tunes a pretrained vision-language model (VLM) using Chain-of-Thought annotated demonstrations, followed by reinforcement learning under multi-level reward signals. Our extensive experiments show that VIKI-R significantly outperforms baselines method across all task levels. Furthermore, we show that reinforcement learning enables the emergence of compositional cooperation patterns among heterogeneous agents. Together, VIKI-Bench and VIKI-R offer a unified testbed and method for advancing multi-agent, visual-driven cooperation in embodied AI systems.
Articulated objects are prevalent in daily life. Understanding their kinematic structure and reconstructing them have numerous applications in embodied AI and robotics. However, current methods require carefully captured data for training or inference, preventing practical, scalable, and generalizable reconstruction of articulated objects. We focus on reconstruction of an articulated object from a casually captured RGBD video shot with a hand-held camera. A casually captured video of an interaction with an articulated object is easy to acquire at scale using smartphones. However, this setting is quite challenging, as the object and camera move simultaneously and there are significant occlusions as the person interacts with the object. To tackle these challenges, we introduce a coarse-to-fine framework that infers joint parameters and segments movable parts of the object from a dynamic RGBD video. To evaluate our method under this new setting, we build a 20$\times$ larger synthetic dataset of 784 videos containing 284 objects across 11 categories. We compare our approach with existing methods that also take video as input. Experiments show that our method can reconstruct synthetic and real articulated objects across different categories from dynamic RGBD videos, outperforming existing methods significantly.
Large-scale video generative models can synthesize diverse and realistic visual content for dynamic world creation, but they often lack element-wise controllability, hindering their use in editing scenes and training embodied AI agents. We propose Dreamland, a hybrid world generation framework combining the granular control of a physics-based simulator and the photorealistic content output of large-scale pretrained generative models. In particular, we design a layered world abstraction that encodes both pixel-level and object-level semantics and geometry as an intermediate representation to bridge the simulator and the generative model. This approach enhances controllability, minimizes adaptation cost through early alignment with real-world distributions, and supports off-the-shelf use of existing and future pretrained generative models. We further construct a D3Sim dataset to facilitate the training and evaluation of hybrid generation pipelines. Experiments demonstrate that Dreamland outperforms existing baselines with 50.8% improved image quality, 17.9% stronger controllability, and has great potential to enhance embodied agent training. Code and data will be made available.
Category-level 3D/6D pose estimation is a crucial step towards comprehensive 3D scene understanding, which would enable a broad range of applications in robotics and embodied AI. Recent works explored neural mesh models that approach a range of 2D and 3D tasks from an analysis-by-synthesis perspective. Despite the largely enhanced robustness to partial occlusion and domain shifts, these methods depended heavily on 3D annotations for part-contrastive learning, which confines them to a narrow set of categories and hinders efficient scaling. In this work, we present DINeMo, a novel neural mesh model that is trained with no 3D annotations by leveraging pseudo-correspondence obtained from large visual foundation models. We adopt a bidirectional pseudo-correspondence generation method, which produce pseudo correspondence utilize both local appearance features and global context information. Experimental results on car datasets demonstrate that our DINeMo outperforms previous zero- and few-shot 3D pose estimation by a wide margin, narrowing the gap with fully-supervised methods by 67.3%. Our DINeMo also scales effectively and efficiently when incorporating more unlabeled images during training, which demonstrate the advantages over supervised learning methods that rely on 3D annotations. Our project page is available at https://analysis-by-synthesis.github.io/DINeMo/.
Automatic indoor layout generation has attracted increasing attention due to its potential in interior design, virtual environment construction, and embodied AI. Existing methods fall into two categories: prompt-driven approaches that leverage proprietary LLM services (e.g., GPT APIs) and learning-based methods trained on layout data upon diffusion-based models. Prompt-driven methods often suffer from spatial inconsistency and high computational costs, while learning-based methods are typically constrained by coarse relational graphs and limited datasets, restricting their generalization to diverse room categories. In this paper, we revisit LLM-based indoor layout generation and present 3D-SynthPlace, a large-scale dataset that combines synthetic layouts generated via a 'GPT synthesize, Human inspect' pipeline, upgraded from the 3D-Front dataset. 3D-SynthPlace contains nearly 17,000 scenes, covering four common room types -- bedroom, living room, kitchen, and bathroom -- enriched with diverse objects and high-level spatial annotations. We further introduce OptiScene, a strong open-source LLM optimized for indoor layout generation, fine-tuned based on our 3D-SynthPlace dataset through our two-stage training. For the warum-up stage I, we adopt supervised fine-tuning (SFT), which is taught to first generate high-level spatial descriptions then conditionally predict concrete object placements. For the reinforcing stage II, to better align the generated layouts with human design preferences, we apply multi-turn direct preference optimization (DPO), which significantly improving layout quality and generation success rates. Extensive experiments demonstrate that OptiScene outperforms traditional prompt-driven and learning-based baselines. Moreover, OptiScene shows promising potential in interactive tasks such as scene editing and robot navigation.
Diffusion models have shown remarkable flexibility for solving inverse problems without task-specific retraining. However, existing approaches such as Manifold Preserving Guided Diffusion (MPGD) apply only a single gradient update per denoising step, limiting restoration fidelity and robustness, especially in embedded or out-of-distribution settings. In this work, we introduce a multistep optimization strategy within each denoising timestep, significantly enhancing image quality, perceptual accuracy, and generalization. Our experiments on super-resolution and Gaussian deblurring demonstrate that increasing the number of gradient updates per step improves LPIPS and PSNR with minimal latency overhead. Notably, we validate this approach on a Jetson Orin Nano using degraded ImageNet and a UAV dataset, showing that MPGD, originally trained on face datasets, generalizes effectively to natural and aerial scenes. Our findings highlight MPGD's potential as a lightweight, plug-and-play restoration module for real-time visual perception in embodied AI agents such as drones and mobile robots.
Embodied AI robots have the potential to fundamentally improve the way human beings live and manufacture. Continued progress in the burgeoning field of using large language models to control robots depends critically on an efficient computing substrate, and this trend is strongly evident in manipulation tasks. In particular, today's computing systems for embodied AI robots for manipulation tasks are designed purely based on the interest of algorithm developers, where robot actions are divided into a discrete frame basis. Such an execution pipeline creates high latency and energy consumption. This paper proposes \textsc{Corki}\xspace, an algorithm-architecture co-design framework for real-time embodied AI-powered robotic manipulation applications. We aim to decouple LLM inference, robotic control, and data communication in the embodied AI robots' compute pipeline. Instead of predicting action for one single frame, \textsc{Corki}\xspace predicts the trajectory for the near future to reduce the frequency of LLM inference. The algorithm is coupled with a hardware that accelerates transforming trajectory into actual torque signals used to control robots and an execution pipeline that parallels data communication with computation. \textsc{Corki}\xspace largely reduces LLM inference frequency by up to $5.1\times$, resulting in up to $5.9\times$ speed up. The success rate improvement can be up to 13.9\%.
Understanding relationships between objects is central to visual intelligence, with applications in embodied AI, assistive systems, and scene understanding. Yet, most visual relationship detection (VRD) models rely on a fixed predicate set, limiting their generalization to novel interactions. A key challenge is the inability to visually ground semantically plausible, but unannotated, relationships hypothesized from external knowledge. This work introduces an iterative visual grounding framework that leverages large language models (LLMs) as structured relational priors. Inspired by expectation-maximization (EM), our method alternates between generating candidate scene graphs from detected objects using an LLM (expectation) and training a visual model to align these hypotheses with perceptual evidence (maximization). This process bootstraps relational understanding beyond annotated data and enables generalization to unseen predicates. Additionally, we introduce a new benchmark for open-world VRD on Visual Genome with 21 held-out predicates and evaluate under three settings: seen, unseen, and mixed. Our model outperforms LLM-only, few-shot, and debiased baselines, achieving mean recall (mR@50) of 15.9, 13.1, and 11.7 on predicate classification on these three sets. These results highlight the promise of grounded LLM priors for scalable open-world visual understanding.
Realistic 3D indoor scene synthesis is vital for embodied AI and digital content creation. It can be naturally divided into two subtasks: object generation and layout generation. While recent generative models have significantly advanced object-level quality and controllability, layout generation remains challenging due to limited datasets. Existing methods either overfit to these datasets or rely on predefined constraints to optimize numerical layout that sacrifice flexibility. As a result, they fail to generate scenes that are both open-vocabulary and aligned with fine-grained user instructions. We introduce DirectLayout, a framework that directly generates numerical 3D layouts from text descriptions using generalizable spatial reasoning of large language models (LLMs). DirectLayout decomposes the generation into three stages: producing a Bird's-Eye View (BEV) layout, lifting it into 3D space, and refining object placements. To enable explicit spatial reasoning and help the model grasp basic principles of object placement, we employ Chain-of-Thought (CoT) Activation based on the 3D-Front dataset. Additionally, we design CoT-Grounded Generative Layout Reward to enhance generalization and spatial planning. During inference, DirectLayout addresses asset-layout mismatches via Iterative Asset-Layout Alignment through in-context learning. Extensive experiments demonstrate that DirectLayout achieves impressive semantic consistency, generalization and physical plausibility.
Reconstructing semantic-aware 3D scenes from sparse views is a challenging yet essential research direction, driven by the demands of emerging applications such as virtual reality and embodied AI. Existing per-scene optimization methods require dense input views and incur high computational costs, while generalizable approaches often struggle to reconstruct regions outside the input view cone. In this paper, we propose OGGSplat, an open Gaussian growing method that expands the field-of-view in generalizable 3D reconstruction. Our key insight is that the semantic attributes of open Gaussians provide strong priors for image extrapolation, enabling both semantic consistency and visual plausibility. Specifically, once open Gaussians are initialized from sparse views, we introduce an RGB-semantic consistent inpainting module applied to selected rendered views. This module enforces bidirectional control between an image diffusion model and a semantic diffusion model. The inpainted regions are then lifted back into 3D space for efficient and progressive Gaussian parameter optimization. To evaluate our method, we establish a Gaussian Outpainting (GO) benchmark that assesses both semantic and generative quality of reconstructed open-vocabulary scenes. OGGSplat also demonstrates promising semantic-aware scene reconstruction capabilities when provided with two view images captured directly from a smartphone camera.
Embodied AI systems, comprising AI models and physical plants, are increasingly prevalent across various applications. Due to the rarity of system failures, ensuring their safety in complex operating environments remains a major challenge, which severely hinders their large-scale deployment in safety-critical domains, such as autonomous vehicles, medical devices, and robotics. While achieving provable deterministic safety--verifying system safety across all possible scenarios--remains theoretically ideal, the rarity and complexity of corner cases make this approach impractical for scalable embodied AI systems. To address this challenge, we introduce provable probabilistic safety, which aims to ensure that the residual risk of large-scale deployment remains below a predefined threshold. Instead of attempting exhaustive safety proof across all corner cases, this paradigm establishes a probabilistic safety boundary on overall system performance, leveraging statistical methods to enhance feasibility and scalability. A well-defined probabilistic safety boundary enables embodied AI systems to be deployed at scale while allowing for continuous refinement of safety guarantees. Our work focuses on three core questions: what is provable probabilistic safety, how to prove the probabilistic safety, and how to achieve the provable probabilistic safety. By bridging the gap between theoretical safety assurance and practical deployment, our work offers a pathway toward safer, large-scale adoption of embodied AI systems in safety-critical applications.
This paper introduces GEX, an innovative low-cost dexterous manipulation system that combines the GX11 tri-finger anthropomorphic hand (11 DoF) with the EX12 tri-finger exoskeleton glove (12 DoF), forming a closed-loop teleoperation framework through kinematic retargeting for high-fidelity control. Both components employ modular 3D-printed finger designs, achieving ultra-low manufacturing costs while maintaining full actuation capabilities. Departing from conventional tendon-driven or underactuated approaches, our electromechanical system integrates independent joint motors across all 23 DoF, ensuring complete state observability and accurate kinematic modeling. This full-actuation architecture enables precise bidirectional kinematic calculations, substantially enhancing kinematic retargeting fidelity between the exoskeleton and robotic hand. The proposed system bridges the cost-performance gap in dexterous manipulation research, providing an accessible platform for acquiring high-quality demonstration data to advance embodied AI and dexterous robotic skill transfer learning.
Open-vocabulary semantic segmentation (OVSS) involves assigning labels to each pixel in an image based on textual descriptions, leveraging world models like CLIP. However, they encounter significant challenges in cross-domain generalization, hindering their practical efficacy in real-world applications. Embodied AI systems are transforming autonomous navigation for ground vehicles and drones by enhancing their perception abilities, and in this study, we present AetherVision-Bench, a benchmark for multi-angle segmentation across aerial, and ground perspectives, which facilitates an extensive evaluation of performance across different viewing angles and sensor modalities. We assess state-of-the-art OVSS models on the proposed benchmark and investigate the key factors that impact the performance of zero-shot transfer models. Our work pioneers the creation of a robustness benchmark, offering valuable insights and establishing a foundation for future research.
While theory and practice are often seen as separate domains, this article shows that theoretical insight is essential for overcoming real-world engineering barriers. We begin with a practical challenge: training a cross-morphology embodied AI policy that generalizes across diverse robot morphologies. We formalize this as the Heterogeneous Embodied Agent Training (HEAT) problem and prove it reduces to a structured Partially Observable Markov Decision Process (POMDP) that is PSPACE-complete. This result explains why current reinforcement learning pipelines break down under morphological diversity, due to sequential training constraints, memory-policy coupling, and data incompatibility. We further explore Collective Adaptation, a distributed learning alternative inspired by biological systems. Though NEXP-complete in theory, it offers meaningful scalability and deployment benefits in practice. This work illustrates how computational theory can illuminate system design trade-offs and guide the development of more robust, scalable embodied AI. For practitioners and researchers to explore this problem, the implementation code of this work has been made publicly available at https://github.com/airs-admin/HEAT
Object goal navigation is a fundamental task in embodied AI, where an agent is instructed to locate a target object in an unexplored environment. Traditional learning-based methods rely heavily on large-scale annotated data or require extensive interaction with the environment in a reinforcement learning setting, often failing to generalize to novel environments and limiting scalability. To overcome these challenges, we explore a zero-shot setting where the agent operates without task-specific training, enabling more scalable and adaptable solution. Recent advances in Vision Foundation Models (VFMs) offer powerful capabilities for visual understanding and reasoning, making them ideal for agents to comprehend scenes, identify relevant regions, and infer the likely locations of objects. In this work, we present a zero-shot object goal navigation framework that integrates the perceptual strength of VFMs with a model-based planner that is capable of long-horizon decision making through frontier exploration. We evaluate our approach on the HM3D dataset using the Habitat simulator and demonstrate that our method achieves state-of-the-art performance in terms of success weighted by path length for zero-shot object goal navigation.
We introduce a new task, Map and Locate, which unifies the traditionally distinct objectives of open-vocabulary segmentation - detecting and segmenting object instances based on natural language queries - and 3D reconstruction, the process of estimating a scene's 3D structure from visual inputs. Specifically, Map and Locate involves generating a point cloud from an unposed video and segmenting object instances based on open-vocabulary queries. This task serves as a critical step toward real-world embodied AI applications and introduces a practical task that bridges reconstruction, recognition and reorganization. To tackle this task, we introduce a simple yet effective baseline, which we denote as SAB3R. Our approach builds upon MASt3R, a recent breakthrough in 3D computer vision, and incorporates a lightweight distillation strategy. This method transfers dense, per-pixel semantic features from 2D vision backbones (eg, CLIP and DINOv2) to enhance MASt3R's capabilities. Without introducing any auxiliary frozen networks, our model generates per-pixel semantic features and constructs cohesive point maps in a single forward pass. Compared to separately deploying MASt3R and CLIP, our unified model, SAB3R, achieves superior performance on the Map and Locate benchmark. Furthermore, we evaluate SAB3R on both 2D semantic segmentation and 3D tasks to comprehensively validate its effectiveness.
Emerging embodied AI applications, such as wearable cameras and autonomous agents, have underscored the need for robust reasoning from first person video streams. We introduce EgoVLM, a vision-language model specifically designed to integrate visual comprehension and spatial-temporal reasoning within egocentric video contexts. EgoVLM is fine-tuned via Group Relative Policy Optimization (GRPO), a reinforcement learning method adapted to align model outputs with human-like reasoning steps. Following DeepSeek R1-Zero's approach, we directly tune using RL without any supervised fine-tuning phase on chain-of-thought (CoT) data. We evaluate EgoVLM on egocentric video question answering benchmarks and show that domain-specific training substantially improves performance over general-purpose VLMs. Our EgoVLM-3B, trained exclusively on non-CoT egocentric data, outperforms the base Qwen2.5-VL 3B and 7B models by 14.33 and 13.87 accuracy points on the EgoSchema benchmark, respectively. By explicitly generating reasoning traces, EgoVLM enhances interpretability, making it well-suited for downstream applications. Furthermore, we introduce a novel keyframe-based reward that incorporates salient frame selection to guide reinforcement learning optimization. This reward formulation opens a promising avenue for future exploration in temporally grounded egocentric reasoning.
While large language models (LLMs) have shown promising capabilities as zero-shot planners for embodied agents, their inability to learn from experience and build persistent mental models limits their robustness in complex open-world environments like Minecraft. We introduce MINDSTORES, an experience-augmented planning framework that enables embodied agents to build and leverage mental models through natural interaction with their environment. Drawing inspiration from how humans construct and refine cognitive mental models, our approach extends existing zero-shot LLM planning by maintaining a database of past experiences that informs future planning iterations. The key innovation is representing accumulated experiences as natural language embeddings of (state, task, plan, outcome) tuples, which can then be efficiently retrieved and reasoned over by an LLM planner to generate insights and guide plan refinement for novel states and tasks. Through extensive experiments in the MineDojo environment, a simulation environment for agents in Minecraft that provides low-level controls for Minecraft, we find that MINDSTORES learns and applies its knowledge significantly better than existing memory-based LLM planners while maintaining the flexibility and generalization benefits of zero-shot approaches, representing an important step toward more capable embodied AI systems that can learn continuously through natural experience.
Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.
Estimating agent pose and 3D scene structure from multi-camera rigs is a central task in embodied AI applications such as autonomous driving. Recent learned approaches such as DUSt3R have shown impressive performance in multiview settings. However, these models treat images as unstructured collections, limiting effectiveness in scenarios where frames are captured from synchronized rigs with known or inferable structure. To this end, we introduce Rig3R, a generalization of prior multiview reconstruction models that incorporates rig structure when available, and learns to infer it when not. Rig3R conditions on optional rig metadata including camera ID, time, and rig poses to develop a rig-aware latent space that remains robust to missing information. It jointly predicts pointmaps and two types of raymaps: a pose raymap relative to a global frame, and a rig raymap relative to a rig-centric frame consistent across time. Rig raymaps allow the model to infer rig structure directly from input images when metadata is missing. Rig3R achieves state-of-the-art performance in 3D reconstruction, camera pose estimation, and rig discovery, outperforming both traditional and learned methods by 17-45% mAA across diverse real-world rig datasets, all in a single forward pass without post-processing or iterative refinement.
We introduce a new task, Map and Locate, which unifies the traditionally distinct objectives of open-vocabulary segmentation - detecting and segmenting object instances based on natural language queries - and 3D reconstruction, the process of estimating a scene's 3D structure from visual inputs. Specifically, Map and Locate involves generating a point cloud from an unposed video and segmenting object instances based on open-vocabulary queries. This task serves as a critical step toward real-world embodied AI applications and introduces a practical task that bridges reconstruction, recognition and reorganization. To tackle this task, we introduce a simple yet effective baseline, which we denote as SAB3R. Our approach builds upon MASt3R, a recent breakthrough in 3D computer vision, and incorporates a lightweight distillation strategy. This method transfers dense, per-pixel semantic features from 2D vision backbones (eg, CLIP and DINOv2) to enhance MASt3R's capabilities. Without introducing any auxiliary frozen networks, our model generates per-pixel semantic features and constructs cohesive point maps in a single forward pass. Compared to separately deploying MASt3R and CLIP, our unified model, SAB3R, achieves superior performance on the Map and Locate benchmark. Furthermore, we evaluate SAB3R on both 2D semantic segmentation and 3D tasks to comprehensively validate its effectiveness.
To enhance immersive experiences, binaural audio offers spatial awareness of sounding objects in AR, VR, and embodied AI applications. While existing audio spatialization methods can generally map any available monaural audio to binaural audio signals, they often lack the flexible and interactive control needed in complex multi-object user-interactive environments. To address this, we propose a Text-guided Audio Spatialization (TAS) framework that utilizes flexible text prompts and evaluates our model from unified generation and comprehension perspectives. Due to the limited availability of premium and large-scale stereo data, we construct the SpatialTAS dataset, which encompasses 376,000 simulated binaural audio samples to facilitate the training of our model. Our model learns binaural differences guided by 3D spatial location and relative position prompts, augmented by flipped-channel audio. It outperforms existing methods on both simulated and real-recorded datasets, demonstrating superior generalization and accuracy. Besides, we develop an assessment model based on Llama-3.1-8B, which evaluates the spatial semantic coherence between our generated binaural audio and text prompts through a spatial reasoning task. Results demonstrate that text prompts provide flexible and interactive control to generate binaural audio with excellent quality and semantic consistency in spatial locations. Dataset is available at \href{https://github.com/Alice01010101/TASU}