Quasi-zero stiffness (QZS) structures are widely employed due to their high static–low dynamic characteristic. While most current QZS structures demonstrate excellent performance in low-frequency vibration isolation, their complex design forms and challenging parameter tuning often limit engineering applications. To address this, this paper proposes a novel multi-segment beam-type QZS structure with simplified geometry and tunable properties. The proposed design consists of two symmetrically arranged inclined straight beams connected to a central platform. The structure first employs Timoshenko beam theory combined with the chain beam constraint method (CBCM) to establish its geometric-mechanical model. Key parameters are then optimized through a Kriging surrogate model and genetic algorithm, enabling QZS over a wide displacement range with flexible parameter-driven adjustability. The structure demonstrates strong global stability and is validated through theoretical analysis and experimental testing. A 3D-printed prototype confirms the vibration isolator’s effective vibration isolation performance, offering a practical solution for compact, high-performance QZS systems in engineering applications.
World models have demonstrated significant promise for data synthesis in autonomous driving. However, existing methods predominantly concentrate on single-modality generation, typically focusing on either multi-camera video or LiDAR sequence synthesis. In this paper, we propose UniDriveDreamer, a single-stage unified multimodal world model for autonomous driving, which directly generates multimodal future observations without relying on intermediate representations or cascaded modules. Our framework introduces a LiDAR-specific variational autoencoder (VAE) designed to encode input LiDAR sequences, alongside a video VAE for multi-camera images. To ensure cross-modal compatibility and training stability, we propose Unified Latent Anchoring (ULA), which explicitly aligns the latent distributions of the two modalities. The aligned features are fused and processed by a diffusion transformer that jointly models their geometric correspondence and temporal evolution. Additionally, structured scene layout information is projected per modality as a conditioning signal to guide the synthesis. Extensive experiments demonstrate that UniDriveDreamer outperforms previous state-of-the-art methods in both video and LiDAR generation, while also yielding measurable improvements in downstream driving tasks.
Accurately predicting battery voltage holds significant value in real-world applications. Restricted to the intrinsic conflict between accuracy and complexity, existing predictive models struggle to achieve precise voltage prediction with low complexity, well protocol compatibility and favorable aging adaptability. To fill this research gap, here we introduce a lightweight machine learning framework for accurate voltage prediction across diverse fast-charging protocols over entire lifespan. Base-error joint modeling, the core mechanism of our method, enables rapid framework construction via extremely partial data, empowering remarkable prediction performance with consideration of contrasting protocols and various degradation. The incorporation with pseudo initial charging voltage generation and required voltage fragment location further endows sufficient flexibility of our framework over arbitrarily partial charging. Our validation, derived from a public battery degradation dataset comprising 8 different multi-step constant current fast-charging protocols and 71,903,990 samples from 46 cells, showcases that our framework can be well trained within 3.3 s, with root mean square error and mean absolute percentage error of voltage prediction less than 52 mV and 1.19% respectively. We expect this framework could also illuminate multi-scenario state-of-charge estimation, aging monitoring and fault diagnosis in the next-generation battery management system.
Qualitative results and an illustration of our core idea. Top left: reconstruction results on benchmark images. Top right: reconstruction results on real-world images. Bottom: illustration of reconstruction-guided noise initialization and modulation. Given multiple input images, we predict a point cloud in canonical space, deterministically inject the predicted geometry into the diffusion process through noise inversion, and modulate the resulting noise to preserve the generative flexibility required to complete unobserved regions and refine visible geometry.
