Recent advances in multimodal large language models (MLLMs) have shown remarkable capabilitiesin integrating vision and language for complex reasoning. While most existing benchmarks evaluate models under offline settings with a fixed set of pre-recorded inputs, we introduce OST-Bench, a benchmark designed to evaluate Online Spatio-Temporal understanding from the perspective of an agent actively exploring a scene. The “Online” aspect emphasizes the need to process and reason over incrementally acquired observations, while the “Spatio-Temporal” component requires integrating current visual inputs with historical memory to support dynamic spatial reasoning. OST-Bench better reflects the challenges of real-world embodied perception. Built on an efficient data collection pipeline, OST-Bench consists of 1.4k scenes and 10k question-answer pairs collected from ScanNet, Matterport3D, and ARKitScenes. We evaluate several leading MLLMs on OST-Bench and observe that they fall short on tasks requiring complex spatio-temporal reasoning. Under the online setting, their accuracy declines as the exploration horizon extends and the memory grows. Through further experimental analysis, we identify common error patterns across models and find that both complex clue-based spatial reasoning demands and long-term memory retrieval requirements significantly drop model performance along two separate axes, highlighting the core challenges that must be addressed to improve online embodied reasoning. To foster further research and development in the field, our codes, dataset, and benchmark are available at https://github.com/InternRobotics/OST-Bench.

The advancement of Embodied AI heavily relies on large-scale, simulatable 3D scene datasets characterized by scene diversity and realistic layouts.However, existing datasets typically suffer from limitations in data scale or diversity, sanitized layouts lacking small items, and severe object collisions.To address these shortcomings, we introduce \textbf{InternScenes}, a novel large-scale simulatable indoor scene dataset comprising approximately 40,000 diverse scenes by integrating three disparate scene sources, \ie, real-world scans, procedurally generated scenes, and designer-created scenes, including 1.96M 3D objects and covering 15 common scene types and 288 object classes.We particularly preserve massive small items in the scenes, resulting in realistic and complex layouts with an average of 41.5 objects per region.Our comprehensive data processing pipeline ensures simulatability by creating real-to-sim replicas for real-world scans, enhances interactivity by incorporating interactive objects into these scenes, and resolves object collisions by physical simulations.We demonstrate the value of InternScenes with two benchmark applications: scene layout generation and point-goal navigation. Both show the new challenges posed by the complex and realistic layouts. More importantly, InternScenes paves the way for scaling up the model training for both tasks, making the generation and navigation in such complex scenes possible. We commit to open-sourcing the data and benchmarks to benefit the whole community.

Contrastive learning, a discriminative self-learning framework, is one of the most popular representation learning methods which has a wide range of application scenarios. Although relative techniques have been continuously updated in recent years, designing and seeking positive pairs are still inevitable. Just because of the requirement of explicit positive pairs, the utilization of contrastive learning is restricted in dense, multi-modal, and other scenarios where positive pairs are difficult to obtain. To solve this problem, in this paper, we design an auto-pairing mechanism called Implicit Relation Circulation (IRC) for discriminative self-learning frameworks. Its core idea is to conduct a random walk among multiple feature groups we want to contrast but without explicit matchup, which we call the complex task (Task C). By linking the head and tail of the random walk to form a circulation with a simple task (task S) containing easy-obtaining pairs, we can apply cycle consistency as supervision guidance to gradually learn the wanted positive pairs among the random walk of feature groups automatically. We provide several amazing applications of IRC: we can learn 1) effective dense image pixel relations and representation with only image-level pairs; 2) 3D temporal point-level multi-modal point cloud relations and representation; and 3) even image representation with the help of language without off-the-shelf vision-language pairs. As an easy-to-use plug-and-play mechanism, we evaluate its universality and robustness with multiple self-learning algorithms, tasks, and datasets, achieving stable and significant improvements. As an illustrative example, IRC improves the SOTA performance by about 3.0 mIoU on image semantic segmentation, 1.5 mIoU on 3D segmentation, 1.3 mAP on 3D detection, and an average of 1.2 top1 accuracy on image classification with the help of the auto-learned positive pairs. Importantly, these improvements are achieved with little parameter and computation overhead. We hope IRC can provide the community with new insight into discriminative self-learning.

With the emergence of LLMs and their integration with other data modalities, multi-modal 3D perception attracts more attention due to its connectivity to the physical world and makes rapid progress. However, limited by existing datasets, previous works mainly focus on understanding object properties or inter-object spatial relationships in a 3D scene. To tackle this problem, this paper builds the first largest ever multi-modal 3D scene dataset and benchmark with hierarchical grounded language annotations, MMScan. It is constructed based on a top-down logic, from region to object level, from a single target to inter-target relationships, covering holistic aspects of spatial and attribute understanding. The overall pipeline incorporates powerful VLMs via carefully designed prompts to initialize the annotations efficiently and further involve humans' correction in the loop to ensure the annotations are natural, correct, and comprehensive. Built upon existing 3D scanning data, the resulting multi-modal 3D dataset encompasses 1.4M meta-annotated captions on 109k objects and 7.7k regions as well as over 3.04M diverse samples for 3D visual grounding and question-answering benchmarks. We evaluate representative baselines on our benchmarks, analyze their capabilities in different aspects, and showcase the key problems to be addressed in the future. Furthermore, we use this high-quality dataset to train state-of-the-art 3D visual grounding and LLMs and obtain remarkable performance improvement both on existing benchmarks and in-the-wild evaluation.