日产

Humanoid loco-manipulation holds transformative potential for daily service and industrial tasks, yet achieving precise, robust whole-body control with 3D end-effector force interaction remains a major challenge. Prior approaches are often limited to lightweight tasks or quadrupedal/wheeled platforms. To overcome these limitations, we propose FALCON, a dual-agent reinforcement-learning-based framework for robust force-adaptive humanoid loco-manipulation. FALCON decomposes whole-body control into two specialized agents: (1) a lower-body agent ensuring stable locomotion under external force disturbances, and (2) an upper-body agent precisely tracking end-effector positions with implicit adaptive force compensation. These two agents are jointly trained in simulation with a force curriculum that progressively escalates the magnitude of external force exerted on the end effector while respecting torque limits. Experiments demonstrate that, compared to the baselines, FALCON achieves 2x more accurate upper-body joint tracking, while maintaining robust locomotion under force disturbances and achieving faster training convergence. Moreover, FALCON enables policy training without embodiment-specific reward or curriculum tuning. Using the same training setup, we obtain policies that are deployed across multiple humanoids, enabling forceful loco-manipulation tasks such as transporting payloads (0-20N force), cart-pulling (0-100N), and door-opening (0-40N) in the real world.

Porosity defects in high-pressure die-casting (HPDC) are a significant quality concern and are challenging to diagnose due to noisy raw data, nonlinear relationships, and data imbalance. Augmenting rare faults becomes essential for machine learning (ML) results, but it can be challenging, potentially leading to the generation of unrealistic or irrelevant fault signatures. This paper considers data augmentation in an explainable AI framework for robust porosity diagnosis. Specifically, we utilize an autoencoder with a Wasserstein generative adversarial network (WGAN-GP) to augment the latent space for rare-fault data and employ the Fréchet Inception Distance (FID) metric to assess the quality of the synthetic data. Demonstrated through a real-world case study. We applied a traditional XGBoost classifier to isolate defects and achieved a 95.90

PurposeThe purpose of this study is to explore novel rotor shapes of variable leakage flux interior permanent magnet synchronous motors (VLF-IPMSMs) via parameter and topology optimization (PTO) to enhance the variable flux characteristics, which are crucial for achieving high torque at low speeds and high efficiency at high speeds in electric vehicle drives.Design/methodology/approachThe PTO method combines the normalized Gaussian network ON/OFF method with covariance matrix adaptation evolution strategy, using an objective function that maximizes flux linkage variation and suppresses harmonic components under torque constraints, which are evaluated using the finite element method.FindingsFor an arc-shaped magnet model identical to the reference model, optimization limited to the end of the magnet region yielded a rotor shape that increased the flux linkage variation by 32.7% and reduced the harmonic index of the radial air gap flux density by 75.2% compared to those of the reference while maintaining the maximum torque. Following the shape sensitivity analysis, low contribution regions were retained in the optimal shape. Upon optimization of the magnet shape parameters, the number of high order harmonic components increased, suggesting convergence to local optima.Originality/valueThis study demonstrates the application of PTO to VLF-IPMSMs using a newly defined objective function. The results demonstrate the potential of the PTO to identify rotor shapes with enhanced flux variations while clarifying the challenges of low sensitivity regions and local optima.

Accurate, up-to-date representations of road structures are critical for the safe operation of autonomous vehicles. Existing systems rely either on costly, maintenance-heavy high-definition (HD) maps which compromise safety when outdated, or purely sensor-based online mapping which struggles with long-range reliability and occlusion. Systems incorporating map prior information into online mapping seek to overcome drawbacks of both approaches by combining them in some way. We propose 'Driveline To HD Map' (D2HDMap), an online mapping system that injects a lightweight, non-visible driveline prior to guide the estimation of visible road structures such as lane dividers, road boundaries and crosswalks. This prior incurs less effort to create and update compared to full HD map priors used in other approaches. We also show that training with such a prior can improve generalization at inference time when no prior is available. Ablation studies conducted on the nuScenes and Argoverse 2 dataset demonstrate that models trained using a driveline prior largely retain performance even when priors are not available. On a geographically disjoint split, D2HDMap achieves 44.8 mAP, surpassing recent state-of-the-art. Additionally, noise-aware training substantially increases robustness to realistic localization error.

To accurately predict the shape of an automobile product, such as a crankshaft, produced by hot die forging, a preliminary simulation of the forging process indicated the significant impacts of local and varying friction coefficients between the complex-shaped die and the material. This study identified the friction coefficients through ring compression and tapered plug penetration tests, focusing on regions with high pressure or large contact areas. The results revealed variations in the friction coefficients across different regions. Consequently, the study suggests implementing locally appropriate friction coefficients on specific die surfaces exhibiting conditions akin to those observed in the friction tests. Specifically, a Coulomb’s friction coefficient of 0.14 was assigned to the product shape region of the crankshaft die. Additionally, a friction model transitioning from a Coulomb’s friction coefficient of 0.5 to a shear friction coefficient of 0.6 was applied in the flash region with significant sliding distances. By incorporating these tailored friction conditions into the simulation of hot die forging for crankshaft manufacturing, the study achieves more accurate material flow, die filling, and underfill replication.