Robotics Software Engineer Β· SLAM Β· Autonomous Navigation
Building autonomous mobile robots that actually ship β from sensor fusion to mass production.
I work on the full stack of autonomous mobile robotics β sensor fusion localization, SLAM, navigation, docking, central FSM design, PLC integration, and on-site commissioning.
Over the past three years I've shipped seven+ industrial robot platforms from prototype to mass-production sites, including:
- π HAMR β a 3-ton-class 4-wheel-swerve heavy-duty AMR with Β±2 cm stop precision over a 50,000 γ‘ map, demonstrated live at Roboworld 2025.
- π€ AMMR β a mobile manipulator validated through two field trials (Chungnam Technopark + a Vietnamese factory).
- π§ AMR Platforms - Home cleaning robot Β· Air-Purifier Robot Β· Autonomous wheelchair Β· Heavy-duty lift robot Β· AED-Bot Β· Rudolph-Bot.
On the research side, I just completed an M.S. at SKKU Intelligent Robotics (GPA 4.33/4.5) with a thesis on RaaS-oriented Semantic Task Management for Heterogeneous Multi-Robot Systems β published in JICROS (KCI / Scopus, 2025).
Currently open to full-time roles in autonomous mobile robotics / autonomous driving SW.
3-ton class Β· Β±2 cm stop precision Β· 50,000 γ‘ map Β· Roboworld 2025 live demo
- ROS2 Jazzy Β· 4-wheel independent swerve drive
- Sensor Fusion: Dual 3D LiDAR + IMU + RTK-GNSS + Wheel encoders
- SLAM: GenZ-ICP-based LiDAR Odometry β 3D point-cloud map construction
- Localization: NDT-OMP @ Adaptive covariance, global relocalization
- PLC Integration: Custom Beckhoff ADS bridge synchronizing TwinCAT
Stack: ROS2 Jazzy Β· SLAM Β· Nav2 Β· ICP Β· NDT Β· EKF Β· Docker
π Closed source Β· Built at Caselab. Architecture details available on request.
Two field trials: Technopark + Vietnamese factory mass-production
- ROS2 Humble Β· VDA 5050 + RESTful API-based ACS integration
- 10-state central FSM:
INIT Β· IDLE Β· AUTO Β· MANUAL Β· DOCKING Β· UNDOCKING Β· CHARGING Β· MANIPULATION Β· EMERGENCY Β· STOP - Three mission scenarios:
LOAD/UNLOAD/HOMEend-to-end orchestration - 2-Method Docking (single node, policy-switching): ArUco 6-DoF PD-aligned (camera/marker) & 2D-LiDAR YAW β X β Y sequential alignment
- Beckhoff ADS bridge with charging-station Service / Action orchestration
Stack: ROS2 Humble Β· C++ Β· Python Β· VDA 5050 Β· RESTful API Β· Beckhoff ADS Β· ArUco Β· Docker Β· Git Submodule
π Closed source Β· Built at Caselab. Architecture details available on request.
11-month industry-academic project Β· iRobot Create 3 base Β· End-to-end validated in Isaac Sim and on real robot
- Frontier-based Active Mapping + Cartographer 2D occupancy grid + Nav2
- Voronoi + connected-components room partitioning β topological semantic graph
- Open-vocabulary perception: RGB-D + Grounded SAM (~15-20 furniture classes)
- Room inference: BGE-M3 embedding similarity + rarity weighting + Prime/Key-object priors
- LLM Task Planner: locally fine-tuned Qwen3-1.7B + LoRA (rank=8, Ξ±=32, SFT 3 epochs, lr=1e-4, bf16) for natural-language β structured task decomposition
- 11-state SAN/STPL central FSM with MPPI local controller for dynamic-obstacle avoidance
Stack: ROS2 Humble Β· NVIDIA Isaac Sim Β· Cartographer Β· Nav2 Β· MPPI Β· Grounded SAM Β· BGE-M3 Β· Qwen3-1.7B + LoRA Β· Hugging Face
π€ Industry-Academic with S-corp Β· partial closed-source Β· framework details in JICROS 2025.
TOSM 3-layer model Β· 20-robot heterogeneous fleet Β· Hospital + Office worlds
Proposes the Triplet Ontological Semantic Model (TOSM) β Symbolic / Explicit / Implicit β as a unified representation of robots, tasks, and environments. Four integrated modules: Semantic Map Generator Β· Semantic Task Planner Β· Cost-aware Task Allocator Β· Failure-Aware Adaptive Reallocation, all on a VDA 5050 + MQTT-based Robot Control Server.
Headline results validated on a 20-robot fleet (4Γ Cleaning Β· Delivery Β· Wheelchair Β· Disinfection Β· Patrol):
| Metric | Baseline | Proposed |
|---|---|---|
| Semantic Grounding Accuracy (ambiguous commands) | 35% | 100% |
| Sub-task Generation Success Rate | 95% | 100% |
| Load Balance Index (lower = better) | 0.37 | 0.21 |
| Priority Satisfaction Rate | 66.7% | 100% |
| Reallocation Response Time on robot failure | β | 192 ms (mean) |
Stack: ROS2 Β· Python Β· VDA 5050 Β· MQTT Β· Gazebo Β· Isaac Sim
π Thesis (SKKU, Feb 2026) Β· π JICROS (KCI / Scopus, 2025)
| Project | Period | Highlights |
|---|---|---|
| Home Air-Purifier Robot (Everybot) | 2024.07 β 2024.12 | ROS2 Humble dual-LiDAR Active SLAM, 8-state manager, Docker rollout |
| Map Calibration System (SK magic) | 2024.09 β 2024.12 | OpenCV Hough auto-rotation, Median + Canny + Morph closing |
| Autonomous Wheelchair | 2024.07 β 2024.11 | CAN/Serial motor, dual 2D LiDAR + camera cross-dot localization |
| Heavy-duty Lift Robot (Jeongdo) | 2024.05 β 2024.07 | pyads ADS bridge, node-edge topo-map + Dijkstra |
| Multi Mobile-Robot HW Build (2-person team) | 2024.03 β 2024.06 | Built 3 platforms end-to-end, Wi-Fi mesh fleet |
| Lane-based Cooperative Driving (KSITS 2024) | 2024.08 β 2024.09 | YOLOv8m-seg + 6-state cooperative driving in CARLA |
| AED-Bot (Patent Filed) | 2022.09 β 2023.06 | ST-GCN + AlphaPose cardiac-arrest recognition + EKF SLAM |
| Rudolph-Bot (Indoor Delivery) | 2022.09 β 2022.12 | ROS Melodic SLAM/Nav on a 1M-KRW budget |
- JICROS (KCI / Scopus, 2025) β Multi-Layer Topo-Semantic Map Generation Framework for High-Level Robot Tasks in Indoor Environments
- ICROS 2025 (Poster) β Multi-Layer Map-based Semantic Room Segmentation Framework for Home Environments
- KSITS 2024 (Oral) β Lane-Information-based Vehicle State Decision for Cooperative Driving
- π₯ Innovation Award β National Startup Idea Competition (Intelligent Robotics Innovation Convergence University Consortium, 2023.11)
- π₯ Grand Prize β Creative Capstone Design Competition (Hanyang ERICA, 2023.06)
- π₯ First Prize β AI-X Robotics Startup Idea Competition (Hanyang Intelligent Robotics, 2023.06)
- π₯ Grand Prize β Online Engineering Academic Festival (Hanyang Univ. Engineering, 2021.11)
- π Autonomous mobile robotics β SLAM, localization, navigation, docking
- π€ Mobile manipulation & multi-robot orchestration
- π§ Semantic mapping, open-vocabulary perception, LLM task planning
- π οΈ Sim-to-real validation pipelines (Isaac Sim Β· Gazebo Β· CARLA)
π§ ksw981215@gmail.com
πΌ LinkedIn β linkedin.com/in/seungwon-kwon
π» GitHub β github.com/Kwon-SeungWon
"Engineering autonomy that doesn't break in the wild."