At Marine Thinking Inc., I develop and deploy autonomy capabilities for unmanned surface vehicles, spanning perception, sensor fusion, tracking, planning, navigation, and control. I designed and validated maritime perception and collision-avoidance systems integrating cameras, LiDAR, radar, AIS, GPS, and IMUs, improving collision-avoidance accuracy by approximately 30%.
I extended the navigation platform with modular perception capabilities, including multi-object tracking, semantic segmentation, vision-language models, free-space estimation, and 3D occupancy prediction. I also developed embedded AI systems for NVIDIA Jetson, Raspberry Pi, and Rockchip, applying quantization and pruning to improve real-time performance by up to 30%.
I reduced the onboard perception software image from 2.1 GB to 0.3 GB and built containerized deployment workflows, CI/CD pipelines, and Kubeflow-based model training and monitoring. I validate performance through simulation, lab testing, hardware integration, and on-water trials.
