Case study
prism_loc
Pluggable LiDAR localization for ROS 2: 2D MCL, 3D NDT-MCL, and LiDAR+IMU+RTK-GNSS ESKF fusion over middleware-free C++17 estimator cores, one map→odom contract.
ROS 2C++17MCLESKF
ROS 2 · C++20 · SLAM & NAVIGATION · AI-NATIVE DEV SYSTEMS
Jung Mo Kang
Robotics Software Engineer·6 yrs
Open to conversations about robotics software, ROS 2 architecture, and AI-native engineering practice.
I build autonomous mobile robots, and the engineering system around foundation models that lets them ship safely.
6 yrs
building & field-deploying robot software
30+
AMRs shipped into factory operation
50+
serving robots in real-world service
Robotics tools, built end-to-end
Estimator cores kept free of middleware, ROS 2 wired in only at the edges, and every performance claim backed by tests plus committed evaluation artifacts, not screenshots.
strata
Pluggable lifelong LiDAR mapping for ROS 2: selectable 2D occupancy / 3D voxel backends, Bayesian persistence with FreMEn periodicity, 6-DoF. Sibling of prism_loc.
ROS 2mappinglifelong
compass
Commitment-based passing for social robot navigation: temporally consistent left/right passing decisions with lexicographic safety. Nav2 controller plugin, Gazebo testbed, and paper (arXiv-ready).
Nav2 pluginsocial navigationpaper
se-mppi-nav2
Nav2-native MPPI local controller: online local-minima detection-and-escape plus a dynamic-obstacle CBF safety filter, validated on a 1,200-trial randomized benchmark.
MPPICBFbenchmark
amr_stack
A complete AMR software stack written from scratch: 2D sim, log-odds mapping, scan-matching SLAM, MCL, A*+DWA planning, nav FSM, HRI GUI, all in pure Python plus C++ ROS 2 ports.
from scratchSLAMplanning
robo-ops
Operator console for AMR/AGV fleet control: 30 interactive screens (dashboard, live map, remote control, KPI analytics) as a fully client-side React 19 + TypeScript app.
React 19TypeScriptfleet ops
Production robotics, delivered on-site
Client and employer names withheld. What “on-site” meant here: months of bring-up on the factory floor, crash triage against real hardware, and pilots carried into commercial rollout.
Factory logistics automation: AMR fleet
AMR software engineer, on-site deployment
- Built ROS-based AMR autonomy and the fleet-side ACS bridge protocol (WebSocket) connecting robots to the plant control system.
- Automated mass installation and mapping/logging workflows, making a 30+ robot rollout at an overseas factory repeatable. The protocol carried the project into commercial rollout.
- Lived on-site for months of bring-up: crash triage with gdb, alarm-code and charging-sequence fixes on real hardware.
ROSC++PythonWebSocketon-site
Serving robots at scale
Feature development, maintenance, field debugging
- Maintained and extended C++ autonomy features across 50+ robots serving in restaurants and cafés; multi-robot showcases at industry exhibitions.
- Built a web app for assembly-conformance QA (Node.js/Express + ROS bridge) that sharply cut manual inspection time, and mass-production Ubuntu images for repeatable manufacturing.
ROSC++Node.jsmanufacturing
Autonomous mobility-assistance chair
Bring-up → autonomy → certification
- Brought a rider-carrying autonomous chair from bare hardware to a drivable, obstacle-avoiding system; tuned avoidance path planning for safety around people.
- Passed public-institution certification testing, cleared for real-site service in a public museum setting.
ROSC++motion planningcertification
Airport guide robot + Android SDK
SDK architecture & robot integration
- Designed a Kotlin Android SDK (rosbridge WebSocket, Coroutines/Flow, MVVM) that lets external apps command ROS robots, no CLI required.
- Integrated with a third-party fleet system and shipped service modes (guidance, signage, docent, patrol) plus custom OS images and OTA updates.
KotlinAndroidSDKfleet integration
Autonomous cart PoC
Full robot software, algorithm tuning
- Delivered an indoor autonomous logistics cart pilot in two months on an existing platform: new sensor integration, side-sensing deceleration on turns, site mapping, operator handover.
ROSC++PoCsensors
3D vision sensor applications
Application engineer
- Built the control, logging, and licensing software for an in-house 3D vision sensor: a C# WinForms application plus a C++ encryption/licensing library.
C#C++.NETlicensing
The system around the model
What sets my work apart is not that I use foundation models. It is the engineering system I build around them, proven on a safety-critical robot codebase.
Pipelines, not prompts
An orchestrator coordinates specialized sub-agents with context isolation per task. The unit of work is a structured task with defined inputs, a verification step, and a record, not a chat.
Gates that make AI safe to ship
Every commit is auto-reviewed; merges gate on AI review plus ASan/UBSan/TSan sanitizers in CI. The gates don't care who wrote the code, human or model.
A platform, not a one-off
Custom skills, event hooks, tool integrations, and cross-session memory: a small internal platform with its own standards, repeatable beyond any single project.
Judgment about model limits
Model output is grounded against real-robot behavior before it is trusted; models are selected per task for reliability; my own FM usage is logged and self-audited.
Writing & case studies
Long-form on how I engineer, and a deep dive on one flagship project: architecture, contracts, and the tradeoffs behind them.
FEATURED WRITEUP
How I Engineer with Foundation Models on a Safety-Critical Robot
Not the prompts: the pipeline, the gates, and the judgment. How a multi-year ROS 1 → ROS 2 migration ships fast without losing quality.
Read the article →CASE STUDY
Inside prism_loc: one map→odom contract for three localizers
Why the estimator cores stay middleware-free, how a single map→odom contract lets 2D MCL, 3D NDT-MCL, and ESKF fusion swap in place, and what that costs.
Read the case study →About
I'm a robotics software engineer with six years building and deploying autonomous mobile robots across smart-factory, logistics, mobility-assistance, airport, and food-service settings. I've shipped 30+ AMRs and 50+ serving robots into real-world operation.
I solve at the system level: onboard autonomy, fleet and control-system integration, GUIs, and communication protocols, not just at the code level. Right now I'm leading a multi-year migration of a legacy ROS 1 robot stack to a modular, middleware-agnostic ROS 2 platform in C++20, owning it from architecture through release. Before software, I trained in international trade, and I bring that domain-first habit of thinking to engineering.
What drives me is turning robots from research demos into products that actually work on the floor.
Core stack
C++20ROS 2PythonKotlinSLAM & navigationNav2DDSLinuxDockerCI/CDmulti-agent AI tooling
Languages
Korean (native) · English (professional) · German (B1)