Vision-language-action (VLA) models have demonstrated strong semantic understanding and zero-shot generalization, yet most existing systems assume an accurate low-level controller with hand-crafted action "vocabulary" such as end-effector pose or root velocity. This assumption confines prior work to quasi-static tasks and precludes the agile, whole-body behaviors required by humanoid whole-body control (WBC) tasks. To capture this gap in the literature, we start by introducing the first sim-to-real-ready, vision-language, closed-loop benchmark for humanoid WBC, comprising over 150 tasks from 10 categories. We then propose LeVERB: Latent Vision-Language-Encoded Robot Behavior, a hierarchical latent instruction-following framework for humanoid vision-language WBC, the first of its kind. At the top level, a vision-language policy learns a latent action vocabulary from synthetically rendered kinematic demonstrations; at the low level, a reinforcement-learned WBC policy consumes these latent verbs to generate dynamics-level commands. In our benchmark, LeVERB can zero-shot attain a 80% success rate on simple visual navigation tasks, and 58.5% success rate overall, outperforming naive hierarchical whole-body VLA implementation by 7.8 times.
We propose CARE (Collision Avoidance via Repulsive Estimation), a plug-and-play module that enhances the safety of vision-based navigation without requiring additional range sensors or fine-tuning of pretrained models. While recent foundation models using only RGB inputs have shown strong performance, they often fail to generalize in out-of-distribution (OOD) environments with unseen objects or variations in camera parameters (e.g., field of view, pose, or focal length). Without fine-tuning, these models may generate unsafe trajectories that lead to collisions, requiring costly data collection and retraining. CARE addresses this limitation by seamlessly integrating with any RGB-based navigation system that outputs local trajectories, dynamically adjusting them using repulsive force vectors derived from monocular depth maps. We evaluate CARE by combining it with state-of-the-art vision-based navigation models across multiple robot platforms. CARE consistently reduces collision rates (up to 100%) without sacrificing goal-reaching performance and improves collision-free travel distance by up to 10.7x in exploration tasks.
Recent advancements in Large Language Models (LLMs) and their multimodal extensions (MLLMs) have substantially enhanced machine reasoning across diverse tasks. However, these models predominantly rely on pure text as the medium for both expressing and structuring reasoning, even when visual information is present. In this work, we argue that language may not always be the most natural or effective modality for reasoning, particularly in tasks involving spatial and geometrical information. Motivated by this, we propose a new paradigm, Visual Planning, which enables planning through purely visual representations, independent of text. In this paradigm, planning is executed via sequences of images that encode step-by-step inference in the visual domain, akin to how humans sketch or visualize future actions. We introduce a novel reinforcement learning framework, Visual Planning via Reinforcement Learning (VPRL), empowered by GRPO for post-training large vision models, leading to substantial improvements in planning in a selection of representative visual navigation tasks, FrozenLake, Maze, and MiniBehavior. Our visual planning paradigm outperforms all other planning variants that conduct reasoning in the text-only space. Our results establish Visual Planning as a viable and promising alternative to language-based reasoning, opening new avenues for tasks that benefit from intuitive, image-based inference.
Endoscopic video-based tasks, such as visual navigation and surgical phase recognition, play a crucial role in minimally invasive surgeries by providing real-time assistance. While recent video foundation models have shown promise, their applications are hindered by (1) computational inefficiencies and (2) suboptimal performance caused by limited data for pre-training in endoscopy. To address these issues, we present EndoMamba, a foundation model designed for real-time inference while learning generalized spatiotemporal representations. First, to mitigate computational inefficiencies, we propose the EndoMamba backbone, optimized for real-time inference. Inspired by recent advancements in state space models, EndoMamba integrates Bidirectional Mamba blocks for spatial modeling within individual frames and vanilla Mamba blocks for past-to-present reasoning across the temporal domain. This design enables both strong spatiotemporal modeling and efficient inference in online video streams. Second, we propose a self-supervised hierarchical pre-training diagram to enhance EndoMamba's representation learning using endoscopic videos and incorporating general video domain knowledge. Specifically, our approach combines masked reconstruction with auxiliary supervision, leveraging low-level reconstruction to capture spatial-temporal structures and high-level alignment to transfer broader knowledge from a pretrained general-video domain foundation model. Extensive experiments on four downstream tasks--classification, segmentation, surgical phase recognition, and localization--demonstrate that EndoMamba outperforms existing foundation models and task-specific methods while maintaining real-time inference speed. The source code is available at https://github.com/TianCuteQY/EndoMamba.
Developing broadly generalizable visual navigation policies for robots is a significant challenge, primarily constrained by the availability of large-scale, diverse training data. While curated datasets collected by researchers offer high quality, their limited size restricts policy generalization. To overcome this, we explore leveraging abundant, passively collected data sources, including large volumes of crowd-sourced teleoperation data and unlabeled YouTube videos, despite their potential for lower quality or missing action labels. We propose Model-Based ReAnnotation (MBRA), a framework that utilizes a learned short-horizon, model-based expert model to relabel or generate high-quality actions for these passive datasets. This relabeled data is then distilled into LogoNav, a long-horizon navigation policy conditioned on visual goals or GPS waypoints. We demonstrate that LogoNav, trained using MBRA-processed data, achieves state-of-the-art performance, enabling robust navigation over distances exceeding 300 meters in previously unseen indoor and outdoor environments. Our extensive real-world evaluations, conducted across a fleet of robots (including quadrupeds) in six cities on three continents, validate the policy's ability to generalize and navigate effectively even amidst pedestrians in crowded settings.
To support the Low Altitude Economy (LAE), it is essential to achieve precise localization of unmanned aerial vehicles (UAVs) in urban areas where global positioning system (GPS) signals are unavailable. Vision-based methods offer a viable alternative but face severe bandwidth, memory and processing constraints on lightweight UAVs. Inspired by mammalian spatial cognition, we propose a task-oriented communication framework, where UAVs equipped with multi-camera systems extract compact multi-view features and offload localization tasks to edge servers. We introduce the Orthogonally-constrained Variational Information Bottleneck encoder (O-VIB), which incorporates automatic relevance determination (ARD) to prune non-informative features while enforcing orthogonality to minimize redundancy. This enables efficient and accurate localization with minimal transmission cost. Extensive evaluation on a dedicated LAE UAV dataset shows that O-VIB achieves high-precision localization under stringent bandwidth budgets. Code and dataset will be made publicly available at: github.com/fangzr/TOC-Edge-Aerial.
High-fidelity simulation is essential for robotics research, enabling safe and efficient testing of perception, control, and navigation algorithms. However, achieving both photorealistic rendering and accurate physics modeling remains a challenge. This paper presents a novel simulation framework--the Unreal Robotics Lab (URL) that integrates the Unreal Engine's advanced rendering capabilities with MuJoCo's high-precision physics simulation. Our approach enables realistic robotic perception while maintaining accurate physical interactions, facilitating benchmarking and dataset generation for vision-based robotics applications. The system supports complex environmental effects, such as smoke, fire, and water dynamics, which are critical for evaluating robotic performance under adverse conditions. We benchmark visual navigation and SLAM methods within our framework, demonstrating its utility for testing real-world robustness in controlled yet diverse scenarios. By bridging the gap between physics accuracy and photorealistic rendering, our framework provides a powerful tool for advancing robotics research and sim-to-real transfer.
Ultrasound imaging of the heart (echocardiography) is widely used to diagnose cardiac diseases. However, obtaining an echocardiogram requires an expert sonographer and a high-quality ultrasound imaging device, which are generally only available in hospitals. Recently, AI-based navigation models and algorithms have been used to aid novice sonographers in acquiring the standardized cardiac views necessary to visualize potential disease pathologies. These navigation systems typically rely on directional guidance to predict the necessary rotation of the ultrasound probe. This paper demonstrates a novel AI navigation system that builds on a decision model for identifying the inferior vena cava (IVC) of the heart. The decision model is trained offline using cardiac ultrasound videos and employs binary classification to determine whether the IVC is present in a given ultrasound video. The underlying model integrates a novel localization algorithm that leverages the learned feature representations to annotate the spatial location of the IVC in real-time. Our model demonstrates strong localization performance on traditional high-quality hospital ultrasound videos, as well as impressive zero-shot performance on lower-quality ultrasound videos from a more affordable Butterfly iQ handheld ultrasound machine. This capability facilitates the expansion of ultrasound diagnostics beyond hospital settings. Currently, the guidance system is undergoing clinical trials and is available on the Butterfly iQ app.
Recent advancements in diffusion-based imitation learning, which show impressive performance in modeling multimodal distributions and training stability, have led to substantial progress in various robot learning tasks. In visual navigation, previous diffusion-based policies typically generate action sequences by initiating from denoising Gaussian noise. However, the target action distribution often diverges significantly from Gaussian noise, leading to redundant denoising steps and increased learning complexity. Additionally, the sparsity of effective action distributions makes it challenging for the policy to generate accurate actions without guidance. To address these issues, we propose a novel, unified visual navigation framework leveraging the denoising diffusion bridge models named NaviBridger. This approach enables action generation by initiating from any informative prior actions, enhancing guidance and efficiency in the denoising process. We explore how diffusion bridges can enhance imitation learning in visual navigation tasks and further examine three source policies for generating prior actions. Extensive experiments in both simulated and real-world indoor and outdoor scenarios demonstrate that NaviBridger accelerates policy inference and outperforms the baselines in generating target action sequences. Code is available at https://github.com/hren20/NaiviBridger.
Visual navigation, a fundamental challenge in mobile robotics, demands versatile policies to handle diverse environments. Classical methods leverage geometric solutions to minimize specific costs, offering adaptability to new scenarios but are prone to system errors due to their multi-modular design and reliance on hand-crafted rules. Learning-based methods, while achieving high planning success rates, face difficulties in generalizing to unseen environments beyond the training data and often require extensive training. To address these limitations, we propose a hybrid approach that combines the strengths of learning-based methods and classical approaches for RGB-only visual navigation. Our method first trains a conditional diffusion model on diverse path-RGB observation pairs. During inference, it integrates the gradients of differentiable scene-specific and task-level costs, guiding the diffusion model to generate valid paths that meet the constraints. This approach alleviates the need for retraining, offering a plug-and-play solution. Extensive experiments in both indoor and outdoor settings, across simulated and real-world scenarios, demonstrate zero-shot transfer capability of our approach, achieving higher success rates and fewer collisions compared to baseline methods. Code will be released at https://github.com/SYSU-RoboticsLab/NaviD.
Indoor poultry farms require inspection robots to maintain precise environmental control, which is crucial for preventing the rapid spread of disease and large-scale bird mortality. However, the complex conditions within these facilities, characterized by areas of intense illumination and water accumulation, pose significant challenges. Traditional navigation methods that rely on a single sensor often perform poorly in such environments, resulting in issues like laser drift and inaccuracies in visual navigation line extraction. To overcome these limitations, we propose a novel composite navigation method that integrates both laser and vision technologies. This approach dynamically computes a fused yaw angle based on the real-time reliability of each sensor modality, thereby eliminating the need for physical navigation lines. Experimental validation in actual poultry house environments demonstrates that our method not only resolves the inherent drawbacks of single-sensor systems, but also significantly enhances navigation precision and operational efficiency. As such, it presents a promising solution for improving the performance of inspection robots in complex indoor poultry farming settings.
Supervised learning (SL) and reinforcement learning (RL) are both widely used to train general-purpose agents for complex tasks, yet their generalization capabilities and underlying mechanisms are not yet fully understood. In this paper, we provide a direct comparison between SL and RL in terms of zero-shot generalization. Using the Habitat visual navigation task as a testbed, we evaluate Proximal Policy Optimization (PPO) and Behavior Cloning (BC) agents across two levels of generalization: state-goal pair generalization within seen environments and generalization to unseen environments. Our experiments show that PPO consistently outperforms BC across both zero-shot settings and performance metrics-success rate and SPL. Interestingly, even though additional optimal training data enables BC to match PPO's zero-shot performance in SPL, it still falls significantly behind in success rate. We attribute this to a fundamental difference in how models trained by these algorithms generalize: BC-trained models generalize by imitating successful trajectories, whereas TD-based RL-trained models generalize through combinatorial experience stitching-leveraging fragments of past trajectories (mostly failed ones) to construct solutions for new tasks. This allows RL to efficiently find solutions in vast state space and discover novel strategies beyond the scope of human knowledge. Besides providing empirical evidence and understanding, we also propose practical guidelines for improving the generalization capabilities of RL and SL through algorithm design.
The aim of this work is to develop an approach that enables Unmanned Aerial System (UAS) to efficiently learn to navigate in large-scale urban environments and transfer their acquired expertise to novel environments. To achieve this, we propose a meta-curriculum training scheme. First, meta-training allows the agent to learn a master policy to generalize across tasks. The resulting model is then fine-tuned on the downstream tasks. We organize the training curriculum in a hierarchical manner such that the agent is guided from coarse to fine towards the target task. In addition, we introduce Incremental Self-Adaptive Reinforcement learning (ISAR), an algorithm that combines the ideas of incremental learning and meta-reinforcement learning (MRL). In contrast to traditional reinforcement learning (RL), which focuses on acquiring a policy for a specific task, MRL aims to learn a policy with fast transfer ability to novel tasks. However, the MRL training process is time consuming, whereas our proposed ISAR algorithm achieves faster convergence than the conventional MRL algorithm. We evaluate the proposed methodologies in simulated environments and demonstrate that using this training philosophy in conjunction with the ISAR algorithm significantly improves the convergence speed for navigation in large-scale cities and the adaptation proficiency in novel environments.
ViVa-SAFELAND is an open source software library, aimed to test and evaluate vision-based navigation strategies for aerial vehicles, with special interest in autonomous landing, while complying with legal regulations and people's safety. It consists of a collection of high definition aerial videos, focusing on real unstructured urban scenarios, recording moving obstacles of interest, such as cars and people. Then, an Emulated Aerial Vehicle (EAV) with a virtual moving camera is implemented in order to ``navigate" inside the video, according to high-order commands. ViVa-SAFELAND provides a new, safe, simple and fair comparison baseline to evaluate and compare different visual navigation solutions under the same conditions, and to randomize variables along several trials. It also facilitates the development of autonomous landing and navigation strategies, as well as the generation of image datasets for different training tasks. Moreover, it is useful for training either human of autonomous pilots using deep learning. The effectiveness of the framework for validating vision algorithms is demonstrated through two case studies, detection of moving objects and risk assessment segmentation. To our knowledge, this is the first safe validation framework of its kind, to test and compare visual navigation solution for aerial vehicles, which is a crucial aspect for urban deployment in complex real scenarios.
Progress in Embodied AI has made it possible for end-to-end-trained agents to navigate in photo-realistic environments with high-level reasoning and zero-shot or language-conditioned behavior, but benchmarks are still dominated by simulation. In this work, we focus on the fine-grained behavior of fast-moving real robots and present a large-scale experimental study involving \numepisodes{} navigation episodes in a real environment with a physical robot, where we analyze the type of reasoning emerging from end-to-end training. In particular, we study the presence of realistic dynamics which the agent learned for open-loop forecasting, and their interplay with sensing. We analyze the way the agent uses latent memory to hold elements of the scene structure and information gathered during exploration. We probe the planning capabilities of the agent, and find in its memory evidence for somewhat precise plans over a limited horizon. Furthermore, we show in a post-hoc analysis that the value function learned by the agent relates to long-term planning. Put together, our experiments paint a new picture on how using tools from computer vision and sequential decision making have led to new capabilities in robotics and control. An interactive tool is available at europe.naverlabs.com/research/publications/reasoning-in-visual-navigation-of-end-to-end-trained-agents.
Palm-sized autonomous nano-drones, i.e., sub-50g in weight, recently entered the drone racing scenario, where they are tasked to avoid obstacles and navigate as fast as possible through gates. However, in contrast with their bigger counterparts, i.e., kg-scale drones, nano-drones expose three orders of magnitude less onboard memory and compute power, demanding more efficient and lightweight vision-based pipelines to win the race. This work presents a map-free vision-based (using only a monocular camera) autonomous nano-drone that combines a real-time deep learning gate detection front-end with a classic yet elegant and effective visual servoing control back-end, only relying on onboard resources. Starting from two state-of-the-art tiny deep learning models, we adapt them for our specific task, and after a mixed simulator-real-world training, we integrate and deploy them aboard our nano-drone. Our best-performing pipeline costs of only 24M multiply-accumulate operations per frame, resulting in a closed-loop control performance of 30 Hz, while achieving a gate detection root mean square error of 1.4 pixels, on our ~20k real-world image dataset. In-field experiments highlight the capability of our nano-drone to successfully navigate through 15 gates in 4 min, never crashing and covering a total travel distance of ~100m, with a peak flight speed of 1.9 m/s. Finally, to stress the generalization capability of our system, we also test it in a never-seen-before environment, where it navigates through gates for more than 4 min.
Visual navigation devices require precise calibration to achieve high-precision localization and navigation, which includes camera and attitude calibration. To address the limitations of time-consuming camera calibration and complex attitude adjustment processes, this study presents a collimator-based calibration method and system. Based on the optical characteristics of the collimator, a single-image camera calibration algorithm is introduced. In addition, integrated with the precision adjustment mechanism of the calibration frame, a rotation transfer model between coordinate systems enables efficient attitude calibration. Experimental results demonstrate that the proposed method achieves accuracy and stability comparable to traditional multi-image calibration techniques. Specifically, the re-projection errors are less than 0.1463 pixels, and average attitude angle errors are less than 0.0586 degrees with a standard deviation less than 0.0257 degrees, demonstrating high precision and robustness.
The object goal visual navigation is the task of navigating to a specific target object using egocentric visual observations. Recent end-to-end navigation models based on deep reinforcement learning have achieved remarkable performance in finding and reaching target objects. However, the collision problem of these models during navigation remains unresolved, since the collision is typically neglected when evaluating the success. Although incorporating a negative reward for collision during training appears straightforward, it results in a more conservative policy, thereby limiting the agent's ability to reach targets. In addition, many of these models utilize only RGB observations, further increasing the difficulty of collision avoidance without depth information. To address these limitations, a new concept -- collision-free success is introduced to evaluate the ability of navigation models to find a collision-free path towards the target object. A two-stage training method with collision prediction is proposed to improve the collision-free success rate of the existing navigation models using RGB observations. In the first training stage, the collision prediction module supervises the agent's collision states during exploration to learn to predict the possible collision. In the second stage, leveraging the trained collision prediction, the agent learns to navigate to the target without collision. The experimental results in the AI2-THOR environment demonstrate that the proposed method greatly improves the collision-free success rate of different navigation models and outperforms other comparable collision-avoidance methods.
This paper introduces a learning-based visual planner for agile drone flight in cluttered environments. The proposed planner generates collision-free waypoints in milliseconds, enabling drones to perform agile maneuvers in complex environments without building separate perception, mapping, and planning modules. Learning-based methods, such as behavior cloning (BC) and reinforcement learning (RL), demonstrate promising performance in visual navigation but still face inherent limitations. BC is susceptible to compounding errors due to limited expert imitation, while RL struggles with reward function design and sample inefficiency. To address these limitations, this paper proposes an inverse reinforcement learning (IRL)-based framework for high-speed visual navigation. By leveraging IRL, it is possible to reduce the number of interactions with simulation environments and improve capability to deal with high-dimensional spaces while preserving the robustness of RL policies. A motion primitive-based path planning algorithm collects an expert dataset with privileged map data from diverse environments, ensuring comprehensive scenario coverage. By leveraging both the acquired expert and learner dataset gathered from the agent's interactions with the simulation environments, a robust reward function and policy are learned across diverse states. While the proposed method is trained in a simulation environment only, it can be directly applied to real-world scenarios without additional training or tuning. The performance of the proposed method is validated in both simulation and real-world environments, including forests and various structures. The trained policy achieves an average speed of 7 m/s and a maximum speed of 8.8 m/s in real flight experiments. To the best of our knowledge, this is the first work to successfully apply an IRL framework for high-speed visual navigation of drones.
Recent success in legged robot locomotion is attributed to the integration of reinforcement learning and physical simulators. However, these policies often encounter challenges when deployed in real-world environments due to sim-to-real gaps, as simulators typically fail to replicate visual realism and complex real-world geometry. Moreover, the lack of realistic visual rendering limits the ability of these policies to support high-level tasks requiring RGB-based perception like ego-centric navigation. This paper presents a Real-to-Sim-to-Real framework that generates photorealistic and physically interactive "digital twin" simulation environments for visual navigation and locomotion learning. Our approach leverages 3D Gaussian Splatting (3DGS) based scene reconstruction from multi-view images and integrates these environments into simulations that support ego-centric visual perception and mesh-based physical interactions. To demonstrate its effectiveness, we train a reinforcement learning policy within the simulator to perform a visual goal-tracking task. Extensive experiments show that our framework achieves RGB-only sim-to-real policy transfer. Additionally, our framework facilitates the rapid adaptation of robot policies with effective exploration capability in complex new environments, highlighting its potential for applications in households and factories.