Institution profile

Mitsubishi Electric Research Laboratories

Industry researchasia · jp
Official website
Research library119linked papers
Opportunities0open roles
Selected work

Representative Papers

A Probability-Guided Sampler for Neural Implicit Surface Rendering

Jun 10, 2025European Conference on Computer Vision

To address inefficient ray sampling and insufficient reconstruction fidelity for foreground implicit surfaces in Neural Radiance Fields (NeRF), this paper proposes an adaptive sampling framework explicitly targeting foreground implicit surfaces. The method models a differentiable probability density function (PDF) directly in the image projection space to guide dense ray sampling within regions of interest. Furthermore, it introduces a novel surface reconstruction loss that jointly optimizes the implicit surface and radiance field by integrating near-surface geometric priors with free-space constraints. Crucially, the approach requires no additional supervision or pretraining and consistently improves mainstream NeRF variants. It significantly enhances geometric accuracy and detail fidelity in target regions while reducing redundant sampling overhead. Experiments demonstrate substantial gains across standard metrics—PSNR, SSIM, and LPIPS—particularly in complex scenes.

1 citationsRead paper

Motion Planning for Autonomous Vehicles: When Model Predictive Control Meets Ensemble Kalman Smoothing

Jul 10, 2024American Control Conference

Nonlinear model predictive control (NMPC) for motion planning using neural-network-based vehicle models suffers from high computational cost and poor real-time performance due to the inherent nonconvexity of the underlying optimization problem. Method: This paper reformulates NMPC as a Bayesian estimation problem—its first such formulation—thereby circumventing traditional numerical optimization bottlenecks. We propose an efficient solution framework based on the ensemble Kalman smoother (EnKS), integrating nonlinear dynamical modeling with sequential data assimilation principles. The approach requires neither gradient evaluation nor iterative optimization, drastically reducing computational complexity. Contribution/Results: Simulation results demonstrate a 100×–1000× improvement in planning speed while preserving trajectory accuracy and closed-loop stability. This work establishes a novel paradigm for real-time NMPC leveraging learned vehicle models, enabling practical deployment in safety-critical autonomous driving applications.

1 citationsRead paper
Recent publications

Latest Papers

Memory-Bounded Continuation of Greedy Sampling for Continual Anomaly Detection

Aug 15, 2026

This study addresses the degradation of coreset representativeness in continual anomaly detection under fixed memory constraints by proposing ContCore. The method introduces a bounded-memory greedy continuation mechanism that employs dynamic sampling expansion and coreset merging strategies to preserve sample representativeness within strict memory limits. Theoretically, ContCore guarantees approximation bound quality while effectively mitigating catastrophic forgetting. Empirical evaluations on the MVTecAD and VisA benchmarks demonstrate that ContCore achieves state-of-the-art performance, significantly outperforming existing approaches particularly in online continual learning scenarios. Consequently, this work provides both theoretical foundations and empirical evidence for effective continual anomaly detection in resource-constrained environments.

0 citationsRead paper