Deheng Zhang

Ganlin Zhang, Deheng Zhang, Feichi Lu et al.

NICE-SLAM is a dense visual SLAM system that combines the advantages of neural implicit representations and hierarchical grid-based scene representation. However, the hierarchical grid features are densely stored, leading to memory explosion problems when adapting the framework to large scenes. In our project, we present sparse NICE-SLAM, a sparse SLAM system incorporating the idea of Voxel Hashing into NICE-SLAM framework. Instead of initializing feature grids in the whole space, voxel features near the surface are adaptively added and optimized. Experiments demonstrated that compared to NICE-SLAM algorithm, our approach takes much less memory and achieves comparable reconstruction quality on the same datasets. Our implementation is available at https://github.com/zhangganlin/NICE-SLAM-with-Adaptive-Feature-Grids.

Ganlin Zhang, Deheng Zhang, Longteng Duan et al.

A novel method to control the Spot robot of Boston Dynamics by Hololens 2 is proposed. This method is mainly designed for people with physical disabilities, users can control the robot's movement and robot arm without using their hands. The eye gaze tracking and head motion tracking technologies of Hololens 2 are utilized for sending control commands. The movement of the robot would follow the eye gaze and the robot arm would mimic the pose of the user's head. Through our experiment, our method is comparable with the traditional control method by joystick in both time efficiency and user experience. Demo can be found on our project webpage: https://zhangganlin.github.io/Holo-Spot-Page/index.html

Nedko Savov, Naser Kazemi, Deheng Zhang et al.

World models have recently gained prominence for action-conditioned visual prediction in complex environments. However, relying on only a few recent observations causes them to lose long-term context. Consequently, within a few steps, the generated scenes drift from what was previously observed, undermining temporal coherence. This limitation, common in state-of-the-art world models, which are diffusion-based, stems from the lack of a lasting environment state. To address this problem, we introduce StateSpaceDiffuser, where a diffusion model is enabled to perform long-context tasks by integrating features from a state-space model, representing the entire interaction history. This design restores long-term memory while preserving the high-fidelity synthesis of diffusion models. To rigorously measure temporal consistency, we develop an evaluation protocol that probes a model's ability to reinstantiate seen content in extended rollouts. Comprehensive experiments show that StateSpaceDiffuser significantly outperforms a strong diffusion-only baseline, maintaining a coherent visual context for an order of magnitude more steps. It delivers consistent views in both a 2D maze navigation and a complex 3D environment. These results establish that bringing state-space representations into diffusion models is highly effective in demonstrating both visual details and long-term memory. Project page: https://insait-institute.github.io/StateSpaceDiffuser/.

Deheng Zhang, Yuqian Fu, Runyi Yang et al.

Most existing benchmarks for egocentric vision understanding focus primarily on daytime scenarios, overlooking the low-light conditions that are inevitable in real-world applications. To investigate this gap, we present EgoNight, the first comprehensive benchmark for nighttime egocentric vision, with visual question answering (VQA) as the core task. A key feature of EgoNight is the introduction of day-night aligned videos, which enhance night annotation quality using the daytime data and reveal clear performance gaps between lighting conditions. To achieve this, we collect both synthetic videos rendered by Blender and real-world recordings, ensuring that scenes and actions are visually and temporally aligned. Leveraging these paired videos, we construct EgoNight-VQA, supported by a novel day-augmented night auto-labeling engine and refinement through extensive human verification. Each QA pair is double-checked by annotators for reliability. In total, EgoNight-VQA contains 3658 QA pairs across 90 videos, spanning 12 diverse QA types, with more than 300 hours of human work. Evaluations of state-of-the-art multimodal large language models (MLLMs) reveal substantial performance drops when transferring from day to night, underscoring the challenges of reasoning under low-light conditions. Beyond VQA, EgoNight also introduces two auxiliary tasks, day-night correspondence retrieval and egocentric depth estimation at night, that further explore the boundaries of existing models. We believe EgoNight-VQA provides a strong foundation for advancing application-driven egocentric vision research and for developing models that generalize across illumination domains. All the data and code will be made available upon acceptance.