Started leading the Data Algorithms section at Bosch, covering automatic labeling, synthetic data, data mining, world models, simulation rendering, and cloud deployment.
Research · Product · Leadership
About Me
I lead the Data Algorithms section at Bosch, responsible for automatic labeling, synthetic data, data mining, world models, simulation rendering, and cloud deployment for autonomous-driving data systems. Since joining Bosch in 2023, I have built teams and new data capabilities from 0 to 1.
Before Bosch, I worked at Huawei Noah’s Ark Lab and the Intelligent Automotive Solution business on on-device vision, LiDAR perception, occupancy ground truth, and multi-modal automatic labeling. I received my M.Eng. in Software Engineering from Sichuan University in 2020.
My work spans the complete data loop: mining the right scenarios, producing reliable ground truth, reconstructing editable 3D worlds, generating targeted synthetic data, and closing the loop with learned simulation. I care most about turning research ideas into systems that help autonomous-driving AI learn faster, scale further, and fail safer.
I am interested in collaborations around 3D Gaussian Splatting, reconstruction-based scene generation, closed-loop world models, and scalable data systems for Physical AI.
News
Recent updatesMM-TRELLIS, our point-cloud-guided approach to multi-modal 3D vehicle generation, was accepted to ICRA 2026.
D²GS, a LiDAR-free method for urban 3D Gaussian Splatting, was published at NeurIPS 2025.
Became Data Closed-loop Algorithm Product Owner and Perception2 Team Leader, leading a five-person team from prototype to production delivery.
Joined Bosch as a Perception Algorithm Expert; delivered production LiDAR perception for parking and placed second in the Waymo 3D Challenge.
Selected Work
Career impactSection leadership · Bosch · 2026—Now
Owning the Data Algorithms portfolio
I lead the section responsible for automatic labeling, synthetic data, data mining, world models, simulation rendering, and cloud deployment. My current focus is a one-stage ADAS evaluation environment built with 3D Gaussian Splatting and world models, while keeping production labeling pipelines reliable at scale.
- 0→1team and capabilities
- 6 domainsend-to-end ownership
- 3DGS + WMevaluation platform
Product ownership · Bosch · 2024—2026
Taking data production from PoC to business delivery
As Data Closed-loop Algorithm Product Owner and Perception2 Team Leader, I led a five-person team from prototype to an end-to-end production workflow and business delivery. We shipped nine automatic-labeling modules at 2,000 clips per day, then built the synthetic-data business from zero and delivered nine projects totaling 3,000 clips and 140,000 frames.
- 5 peopleteam leadership
- 2,000/dayproduction capacity
- +300%pre-labeling throughput
Business outcomes included replacing real AEB data with synthetic data to reduce cost by 90%, and improving general curb performance by 11.2 points in construction scenarios.
Production perception · Bosch · 2023—2024
Shipping LiDAR perception for automated parking
I delivered production dynamic-object perception for parking, developed a multi-task and multi-source framework spanning detection and segmentation, and helped translate research performance into an efficient vehicle-ready stack.
- 75% mAPproduction perception
- #2Waymo 3D Challenge
- +30%inference speed
Algorithm engineering · Huawei · 2020—2023
Building the perception and ground-truth foundation
I moved from on-device ADAS and automated-parking vision to cloud perception at Noah’s Ark Lab. My work covered traffic-sign and pedestrian detection, small-obstacle detection, parking-space delivery, stacked-frame occupancy ground truth, scene flow, tracking, and LiDAR-camera-map fusion.
- +21 mAPtraffic-sign detection
- 99% / 95%parking-space precision / recall
- +3.9 / +1.5labeling precision / recall
Experience
Industry and educationSection Manager, Data Algorithms
Bosch
Leading automatic labeling, synthetic data, data mining, world models, simulation rendering, and cloud deployment. Building one-stage evaluation rendering with 3DGS and world models to reduce ADAS testing cost.
Data Closed-loop Algorithm PO & Perception2 Team Leader
Bosch
Led a five-person team from PoC to production across automatic labeling and synthetic data. Delivered nine algorithm modules at 2,000 clips/day; improved pre-labeling throughput by 300% and delivered nine synthetic-data projects.
Perception Algorithm Expert
Bosch
Delivered production LiDAR dynamic perception for parking at 75% mAP, placed second in the Waymo 3D Challenge, and improved multi-task inference speed by 30% with a two-point mAP gain.
Cloud Perception Algorithm Engineer
Huawei Noah’s Ark Lab · Autonomous Driving
Built occupancy ground truth and multi-modal automatic-labeling modules spanning detection, tracking, scene flow, and LiDAR-camera-map fusion.
Perception Algorithm Engineer
Huawei Intelligent Automotive Solution BU
Developed vision perception for parking and ADAS. Improved traffic-sign mAP by 21 points and pedestrian mAP by 11.2 points; delivered parking-slot detection at 99% precision and 95% recall.
M.Eng. in Software Engineering
Sichuan University · Chengdu
Research in autonomous driving and computer vision, including monocular semantic mapping and extremely low-resolution action recognition.
Selected Publications
Full list on Google Scholar ↗MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving
H. Xiao, Y. Zhang, Y. Bai, et al.
ICRA 2026
D²GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction
K. Xia, J. Jia, K. Jin, Y. Bai, et al.
NeurIPS 2025
Extremely Low Resolution Action Recognition with Confident Spatial-Temporal Attention Transfer
Y. Bai, Q. Zou, et al.
IJCV
Low-Dose CT Image Denoising Using Residual Convolutional Network with Fractional TV Loss
M. Chen, Y.-F. Pu, Y.-C. Bai
Neurocomputing
Monocular Outdoor Semantic Mapping with a Multi-task Network
Y. Bai, L. Fan, Z. Pan, L. Chen
IROS
A Fractional Total Variational CNN Approach for SAR Image Despeckling
Y. Bai, S. Zhang, M. Chen, et al.
ICIC
Contact
I am always glad to exchange ideas about autonomous-driving data, neural reconstruction, synthetic worlds, and production AI systems.
raymondbyc@gmail.com ↗