Social Navigation
Policies and planners for safe, efficient, socially compliant motion in dynamic crowds.
hongson@research:~$
./run_social_intelligence
PERCEPTION · PREDICTION · NAVIGATION · LEARNING
Building robots that move with social intelligence.
I research human-aware autonomy for mobile robots — from modeling pedestrian interactions and forecasting trajectories to learning navigation policies that operate safely around people.
AFFILIATION
NCKU
LAB
NRSL
FOCUS
SOCIAL ROBOTICS
LOCAL FRAME
HUMAN-AWARE NAV
LIVE
MIN DIST—
TTCSAFE
SPEED0.00
LOOP30Hz
> nav_state:
NOMINAL
hongson@nav:~$
SFM
RL
SSM
SCROLL TO EXPLORE
01 / 06
ROS 2✦ISAACLAB✦PYTORCH✦
SOCIAL NAVIGATION✦TRAJECTORY PREDICTION✦
REINFORCEMENT LEARNING✦MAMBA / SSM✦
GRAPH LEARNING✦LiDAR✦HRI✦
ROS 2✦ISAACLAB✦PYTORCH✦
SOCIAL NAVIGATION✦TRAJECTORY PREDICTION✦
REINFORCEMENT LEARNING✦MAMBA / SSM✦
GRAPH LEARNING✦LiDAR✦HRI✦
02
// PROFILE
BACKGROUND & EXPERTISE
Researching how robots understand, predict, and adapt around people.
My work sits at the intersection of robotics, machine learning, human motion intelligence, and autonomous navigation.
I am especially interested in representations that encode human state, social interaction structure, group behavior, dynamic occupancy, and uncertainty — and in using those representations to improve downstream planning and reinforcement-learning policies.
“A robot navigating among people should reason about more than free space; it should reason about interaction.”
~/hongson/profile.yaml
name: Hong-Son Nguyen
role: Research Fellow
degree: M.S. Robotics and Control Systems
institution: National Cheng Kung University
lab: Networked Robotic Systems Laboratory
advisor: Prof. Yen-Chen Liu
location: Tainan, Taiwan
prior: B.S. Mechatronic Engineering, HaUI
research:
- social_robot_navigation
- trajectory_prediction
- robot_learning
- human_interaction_modeling
- human_aware_autonomy
status: active_researchTrajectory Prediction
Multi-agent forecasting with dynamic graphs, structured priors and state-space sequence models.
Robot Learning
Reinforcement learning and simulation-scale training for adaptive human-aware navigation.
Human Modeling
Interaction structure, social forces, grouping, human state encoding and dynamic occupancy.
03
// EXPERIENCE.LOG
ACADEMIC JOURNEY
From simulation-scale learning to real robot experiments.
2024 — PRESENT
ACTIVE
TAINAN, TAIWAN
M.S. in Robotics and Control Systems
National Cheng Kung University · Dept. of Mechanical Engineering
Research on human-aware mobile robot navigation, multi-agent trajectory prediction, pedestrian comfort, reinforcement learning, and real-world autonomous systems. Advisor: Prof. Yen-Chen Liu. GPA 3.52/4.0.
RESEARCH
NRSL
NCKU
Research Fellow · Networked Robotic Systems Lab
Advisor: Prof. Yen-Chen Liu
Developing end-to-end robotics pipelines spanning sensing, human-state representation, prediction, planning, learning, control, and physical validation.
ROBOTAgilex Scout Mini
VISIONZED-X
LiDARVLP-16
COMPUTEJetson AGX Orin
2022 — 2024
TEAM LEAD
HANOI, VIETNAM
Research Assistant · Hanoi Intelligent Robotics Lab
Hanoi University of Industry
Led a student research team working on reinforcement learning for autonomous mobile robots, the line of work that produced the first published navigation papers.
2019 — 2023
BEST THESIS
HANOI, VIETNAM
B.S. in Mechatronic Engineering
Hanoi University of Industry · Advisor: Assoc. Prof. Anh-Tu Nguyen
Thesis on an autonomous 18-DOF hexapod robot trained with deep reinforcement learning, graded 94 and selected as best thesis. GPA 3.05/4.0.
THESIS18-DOF Hexapod
METHODDeep RL
AWARD1st Prize · HaUI 2022
MEMBERASME since 2022
04
// RESEARCH_PROJECTS
SELECTED SYSTEMS
Engineering the social intelligence stack.
Click any project to open its research console. Replace placeholder links with your paper, code, video, and project-page URLs.
PRJ-001
RESEARCH
SCANEGO-CENTRICHover a node to inspect its social state.
DYNAMIC INTERACTION GRAPH // SELECTIVE SSM
TRAJECTORY PREDICTION · 2026
Social Graph Mamba
Attention-free multi-agent trajectory prediction using dynamic interaction graphs, social triplet scans, community structure, and selective state-space models.
PRJ-002
ACTIVE
LEVEL 25 HUMANSReward shaping active
MULTI-ENV POLICY TRAINING // ISAACLAB
ROBOT LEARNING · 2026
RL Social Navigation
Large-scale social-navigation training with pedestrians, obstacles, curriculum learning, reward shaping, and configurable spawn/goal distributions.
PRJ-003
STUDY
CONCEPTUALTTC —Adjust speed and passing offset.
HALLWAY ENCOUNTER // HUMAN COMFORT
HUMAN-ROBOT INTERACTION · 2026
Pedestrian Comfort Modeling
Studying how robot speed, passing geometry, curvature, clearance and time-to-collision relate to perceived comfort in hallway encounters.
PRJ-004
REAL ROBOT
→
→
→
PERCEPTION
ZED-X + VLP-16
RGB/person tracking and geometric obstacle sensing feed the robot-centric state.
CLICK PIPELINE STAGES // PERCEPTION → PREDICTION → CONTROL
AUTONOMOUS ROBOTICS · 2025–2026
Human-Aware Navigation System
A physical robotics pipeline integrating human tracking, 3D LiDAR, trajectory prediction, social interaction modeling and autonomous navigation.
05
// PUBLICATIONS_REGISTRY
GOOGLE SCHOLAR · SELECTED RESEARCH
Published research & conference contributions.
Publications indexed on my Google Scholar profile, including preprints released on arXiv. Peer-reviewed venues and preprints are labelled separately, so the review status of each entry stays explicit.
OPEN GOOGLE SCHOLAR ↗PUB-001
Social Graph Mamba: Forecasting Pedestrian Movements Based on Social Context
Pedestrian trajectory forecasting with selective state space models running over dynamically built interaction graphs, replacing the quadratic cost of attention-based social modeling with linear sequence complexity. A community-aware module using differentiable MinCut recovers group structure, and the predictions are validated on ETH/UCY and SDD as well as on a physical robot coupled to a Social Force Model.
arXiv:2608.21411 [cs.RO], submitted 11 August 2026.
State Space Models
Trajectory Forecasting
Interaction Graphs
Social Context
PUB-002
Stability and Comfort in Mobile Robot–Pedestrian Interactions
Navigation algorithms for nonholonomic mobile robots built on the Social Force Model and its time-to-collision variant, with a formal stability proof for the robot–pedestrian interaction. Simulation and experiments with real pedestrians, together with statistical analysis of comfort surveys, show the models stay stable while improving pedestrian comfort in crowds.
arXiv:2607.17604 [cs.RO; eess.SY], submitted 20 July 2026. 21 pages.
Nonholonomic Robots
Social Force Model
Stability Analysis
Pedestrian Comfort
PUB-003
Empirical Prediction of Pedestrian Comfort in Mobile Robot–Pedestrian Encounters
Empirical modeling of subjective pedestrian comfort from robot–pedestrian interaction kinematics, including minimum distance, projected TTC, and a composite comfort predictor.
Human–Robot Interaction
Pedestrian Comfort
TTC
Mobile Robot Navigation
PUB-004
RL Optimization for Social Force Model in Human-Aware Navigation
Reinforcement-learning-based optimization of Social Force Model parameters for human-aware robot navigation, targeting safer and more adaptive interaction behavior than manually tuned parameters.
Reinforcement Learning
Social Force Model
Human-Aware Navigation
PUB-005
Combination of Deep Reinforcement Learning and Pure-Pursuit Algorithm for Adaptive Autonomous Mobile Robot Navigation
A hybrid autonomous-navigation approach combining deep reinforcement learning with Pure Pursuit path tracking for adaptive mobile-robot motion.
Proceedings of the International Conference on Intelligent Systems and Networks
(ICISN 2024), Lecture Notes in Networks and Systems, Vol. 1077, pp. 563–572.
Deep RL
Pure Pursuit
Path Tracking
Autonomous Navigation
PUB-006
The Application of the Reinforcement Learning Method for Mobile Robot Navigation in an Unknown Environment
Reinforcement-learning-based mobile robot navigation in an unknown environment, with Q-table learning evaluated through Gazebo simulation and experimental validation.
Proceedings of the 3rd Annual International Conference on Material,
Machines and Methods for Sustainable Development (MMMS2022),
Lecture Notes in Mechanical Engineering, pp. 201–207.
Reinforcement Learning
Q-Learning
Gazebo
Unknown Environment
PUB-007
Navigation for Mobile Robot in Unknown Environment using Reinforcement Learning Methods
A navigation framework combining Q-Learning with an obstacle-avoidance method, benchmarked against SARSA in simulation to show the efficiency of the proposed approach.
Journal of Science and Technology, Hanoi University of Industry,
Vol. 59, No. 2B, 2023.
Q-Learning
SARSA
Obstacle Avoidance
Unknown Environment
PREPRINT / UNDER REVIEW
Social Graph Mamba is public as an arXiv preprint and now listed as PUB-001, but it has not yet completed peer review. The implementation and robot integration continue as an active project.
VIEW ONGOING PROJECT ↗
AI// LOCAL_RESEARCH_ASSISTANT
Ask the portfolio.
This lightweight assistant runs entirely in the browser. Try:
research, skills, robot,
publications, contact, or help.
~/hongson/research-assistant
ONLINE
SYS
Research console ready. Ask about Hong-Son's work.
06
// CONTACT_PROTOCOL
OPEN CHANNEL
Let’s build systems that make robots better neighbors.
Open to research conversations, collaborations, PhD opportunities, and work at the intersection of robotics, autonomy, and machine learning.
hongson@portfolio:~$
$ echo "Initialize communication"
Channel ready.
$ open email
hongsonnguyen.haui@gmail.com ↗$ open github
github.com/sontypo ↗$ open linkedin
linkedin.com/in/hong-son-nguyen-751tps/↗$ open scholar
Google Scholar ↗$ download cv --academic
HongSon_Nguyen_CV_Academic.pdf ↓$ download cv --industry
HongSon_Nguyen_CV_Industry.pdf ↓$