Navigation is a fundamental capability for mobile robots. While the current trend is to use learning-based approaches to replace traditional geometry-based methods, existing end-to-end learning-based policies often struggle with 3D spatial reasoning and lack a comprehensive understanding of physical world dynamics. Integrating world models-which predict future observations conditioned on given actions-with iterative optimization planning offers a promising solution due to their capacity for imagination and flexibility. However, current navigation world models, typically built on pure transformer architectures, often rely on multi-step diffusion processes and autoregressive frame-by-frame generation. These mechanisms result in prohibitive computational latency, rendering real-time deployment impossible. To address this bottleneck, we propose a lightweight navigation world model that adopts a one-step generation paradigm and a 3D U-Net backbone equipped with efficient spatial-temporal attention. This design drastically reduces inference latency, enabling high-frequency control while achieving superior predictive performance. We also integrate this model into an optimization-based planning framework utilizing anchor-based initialization to handle multi-modal goal navigation tasks. Extensive closed-loop experiments in both simulation and real-world environments demonstrate our system's superior efficiency and robustness compared to state-of-the-art baselines.
As autonomous vehicle technology continues to evolve, ensuring its safety in complex and dynamic environments has become a critical challenge. Knowledge graphs (KGs) and large language models (LLMs), as two cutting-edge approaches representing the forefront of knowledge-driven and connectionist paradigms in modern artificial intelligence, are emerging as powerful tools for enhancing autonomous vehicle safety. In this study, we provide a comprehensive review of the applications of KGs and LLMs in enhancing the safety of autonomous driving systems. Building upon a brief introduction to the fundamental concepts and underlying technologies of KGs and LLMs, we then present their respective applications in autonomous vehicle safety from complementary perspectives. We further compare the advantages and limitations of KGs and LLMs in terms of knowledge representation, inference capability, scalability, and real-time performance. To leverage the complementary strengths of structured knowledge and language based reasoning, we review existing research efforts that integrate KGs and LLMs in the context of autonomous vehicle (AV) safety enhancement. Based on this analysis, we propose a hybrid safety-enhancement framework that combines explicit knowledge structures of KGs with the flexible reasoning capabilities of LLMs, offering a promising direction toward more robust, interpretable, and adaptive autonomous driving systems.
Imitation learning has demonstrated strong performance in robotic manipulation by learning from large-scale human demonstrations. While existing models excel at single-task learning, it is observed in practical applications that their performance degrades in the multi-task setting, where interference across tasks leads to an averaging effect. To address this issue, we propose to learn diverse skills for behavior models with Mixture of Experts, referred to as Di-BM. Di-BM associates each expert with a distinct observation distribution, enabling experts to specialize in sub-regions of the observation space. Specifically, we employ energy-based models to represent expert-specific observation distributions and jointly train them alongside the corresponding action models. Our approach is plug-and-play and can be seamlessly integrated into standard imitation learning methods. Extensive experiments on multiple real-world robotic manipulation tasks demonstrate that Di-BM significantly outperforms state-of-the-art baselines. Moreover, fine-tuning the pretrained Di-BM on novel tasks exhibits superior data efficiency and the reusable of expert-learned knowledge.
In the case where the color coordinates of the screen display are consistent across different display devices, there can still be inconsistencies in the perceived display effect by the human eye, which is known as the metamerism failure. Especially between OLED displays and LCD displays, due to the significant spectral differences between them, even if they are calibrated to the same color gamut and white point, there are still noticeable differences in their subjective effects. This study proposes a compensation scheme for this problem: selecting one OLED display and one LCD display, choosing 20 different color tones based on the commonly used color distribution of the display, and organizing 11 observers with normal color vision aged 20‐30 to conduct a color matching experiment. Based on the objective chromaticity differences of 20 sets of colors obtained from color matching experiments, perform chromaticity compensation on other colors. The compensation method is to compensate proportionally based on the distance between other colors and 20 sets of calibrated colors in the color space, and reflect it in the form of 3D LUT. The LCD display was colormapped through the updated 3D LUT to further study the reliability and effectiveness of the compensation scheme. The results showed that the compensation scheme could improve the metamerism failure problem between different display devices and improve the consistency of display effects on different devices. In addition, based on CIE 2015 2 ° CMF and color compensation results, the chromaticity calculation of LCD displays was corrected, and proprietary parameters for LCD displays were generated, achieving consistency in subjective display effects and objective chromaticity values between LCD displays and OLED displays.
We developed a way to reduce the effects of these visual effects by multi‐frequency‐de‐burn‐in (MFDBI). The specific scheme of MFDBI is to replace the fixed high‐low refresh rate boundary with dynamic random updating of high‐low refresh rate boundary within a certain number of lines. Through MFDBI, we can not only improve the display effect by blurring the boundary of high and low refresh rate, but also remain a part of the contribution of MFD technology to low power consumption. the brightness difference of the boundary visual effect of the “burning” is reduced
