Yucai Bai

Section Manager · Data Algorithms at Bosch

Building the data
engine for intelligent
vehicles.

I lead teams at the intersection of autonomous-driving data, neural simulation, and 3D reconstruction—turning research ideas into production systems.

Explore selected work raymondbyc@gmail.com
Current focusData flywheelsWorld modelsPhysical AI
Suzhou · ShanghaiCN / EN
01 / Profile
Portrait of Yucai Bai

AI researcher turned engineering leader.

Research. Product. Leadership.

I build systems that make autonomous-driving AI learn faster, scale further, and fail safer.

My work spans the complete data loop: mining the right scenarios, producing reliable ground truth, reconstructing editable 3D worlds, generating synthetic data, and closing the loop with learned simulation.

Today, I lead data-algorithm and simulation initiatives at Bosch. Previously, I worked on perception, automatic labeling, occupancy ground truth, ADAS, and automated parking at Huawei Noah's Ark Lab and the Intelligent Automotive Solution business.

GitHub Google Scholar Email
02 / Focus

From raw reality to
trainable intelligence.

I connect four technical layers into one production-grade data and simulation stack.

01

Data engines

Scenario mining, 4D auto-labeling, quality systems, cloud pipelines, and delivery workflows built for scale.

  • Data closed loop
  • Auto-labeling
  • Mining & quality
02

Neural reconstruction

LiDAR-camera fusion, visual mapping, 3D Gaussian Splatting, and editable digital environments.

  • 3D / 4D mapping
  • 3DGS
  • Vision-only geometry
03

Synthetic data

Reconstruction-based scene editing and generative pipelines for rare, costly, and safety-critical scenarios.

  • Scene synthesis
  • Asset generation
  • Sim2Real evaluation
04

World models

Action-conditioned, temporally coherent sensor simulation for end-to-end closed-loop testing.

  • Closed-loop simulation
  • Long-horizon consistency
  • Physical AI
03 / Selected work

Systems and research
built to move.

A selection of public research and professional themes. Production details are intentionally kept at a non-confidential level.

Conceptual visualization of an urban scene transitioning into a 3D reconstructionNeurIPS 2025
Research2025

D²GS: LiDAR-free urban scene reconstruction

Dense depth regularization and diffusion priors for accurate 3D Gaussian Splatting—reducing dependence on precisely aligned LiDAR while improving urban geometry.

Read the paper
Engineering leadership2023—Now

Autonomous-driving data flywheel

Leading the 0→1 build-out of auto-labeling, synthetic data, data mining, and cloud production capabilities for driving and parking programs.

ICRA · 2026
ICRA · 3D generation2026

MM-TRELLIS: Multi-modal 3D vehicle generation

A point-cloud-guided extension of native 3D generation for high-fidelity vehicle assets from in-the-wild driving data.

Read the paper
Occupancy ground-truth visualization generated from multi-frame point clouds4D perception
Perception data2022—2025

4D automatic labeling and occupancy ground truth

Multi-frame fusion, motion compensation, semantic labeling, tracking, and velocity estimation for precise off-board ground truth.

04 / Journey

A career shaped by
building the next layer.

From perception algorithms to data platforms, and from individual contribution to technical leadership.

2026—Now

Section Manager, Data Algorithms

Bosch

Leading teams across automatic labeling, synthetic data, data mining, world models, simulation rendering, and cloud deployment for autonomous-driving data systems.

LeadershipWorld modelsData platforms
2023—2026

Data Algorithm Lead & Product Owner

Bosch

Industrialized 4D auto-labeling and reconstruction-based synthetic data, established new capabilities from proof of concept to production, and supported large-scale driving and parking programs.

0→1 systems3DGSProduction AI
2020—2023

Algorithm Engineer

Huawei Noah's Ark Lab · IAS BU

Developed perception and data algorithms for ADAS, automated parking, multi-modal automatic labeling, and occupancy ground-truth generation.

PerceptionAuto-labelingOccupancy
2017—2020

M.Eng. in Software Engineering

Sichuan University

Research in computer vision, multi-task learning, semantic 3D reconstruction, and extremely low-resolution action recognition.

Computer visionTransfer learningResearch
2013—2017

B.Eng. in Software Engineering

Chongqing University of Posts and Telecommunications

Foundation in software systems, web engineering, and applied machine learning.

05 / Research

Selected
publications.

Research across 3D reconstruction, autonomous-driving simulation, transfer learning, and visual perception.

2026MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous DrivingH. Xiao, Y. Zhang, Y. Bai, et al.ICRA 2025D²GS: Dense Depth Regularization for LiDAR-free Urban Scene ReconstructionK. Xia, J. Jia, K. Jin, Y. Bai, et al.NeurIPS 2023Extremely Low Resolution Action Recognition with Confident Spatial-Temporal Attention TransferY. Bai, Q. Zou, et al.IJCV 2021Low-Dose CT Image Denoising Using Residual Convolutional Network with Fractional TV LossM. Chen, Y.-F. Pu, Y.-C. BaiNeurocomputing 2019Monocular Outdoor Semantic Mapping with a Multi-task NetworkY. Bai, L. Fan, Z. Pan, L. ChenIROS 2018A Fractional Total Variational CNN Approach for SAR Image DespecklingY. Bai, S. Zhang, M. Chen, et al.ICIC
View Google Scholar profile
06 / Connect

Ideas become systems through good conversations.

Let's build what
comes next.

I'm interested in ambitious work across autonomous-driving data, neural simulation, world models, and Physical AI.

raymondbyc@gmail.com
GitHubGoogle Scholar