I am a Phd student specialized in Computing Science at Simon Fraser University, advised by Dr. Mo Chen (2023 – present).
Check out our lab website: Mars Lab.
I obtained my master degree at University of California, San Diego with a major of Mechanical and Aerospace Engineering in The Safe Autonomous Systems Lab (2021 – 2023).
I worked as a research intern at MaRobot AI in Vancouver on probabilistic human prediction with online intention inference (2025 – 2026).
Previously, I worked as a research assistant in the Institute of Medical Robotics at Shanghai Jiaotong University (2020 – 2021).
I received my B.Eng degree at Tongji University, China majoring in mechanical design, manufacturing and its automation (2015 – 2020).
I was also an exchange student at University of California, Berkeley (2019).
My Research Interest lies in Reachability, Control Theory, Medical Robots, and Generalizable, Socially-aware Human Prediction for Robot Navigation.
Skills
Matlab, Python, C/C++, Java, and LaTeX.
Packages & tools: Jax, HeteroCL, PyTorch, TensorFlow, NumPy, OpenCV, ROS/ROS2, Blender, Gazebo, and CoppeliaSim.
Familliar with mechanical design tools such as AutoCAD and SolidWorks.
Languages: English and Mandarin.
Love skiing and playing video games at leisure time.
- Chong He, Mo Chen, Jacob Daniel Bayless, Michael Samuel Lu, Nhat Minh Bui, "Methods and Systems for Predicting One or More Future States of an Entity", PCT International Patent Application, filed Feb 2026 (pending)
- Chong He, "An Intelligent Shopping Cart", utility model patent, State Intellectual Property Office
Internship Experiences
MaRobot AI (Vancouver, British Columbia, Canada) — Research Intern, 04/2025 - 03/2026
- Developed a probabilistic framework for human prediction with online intention inference
- Implemented the framework in real-world scenarios with ROS2 using Python
- Parallelized code with Jax
Shanghai Jiaotong University – Institute of Medical Robotics — Research Assistant, 06/2020 - 02/2021
- Examined the kinematics of continuum manipulators with visual data
- Constructed experimental platforms for several projects
- Worked with the Robot Operating System (ROS) using C++
Eaton Cooper Electronic Technologies (Shanghai) Co., Ltd. — Engineering Support Intern, Mech&DCC, 08/2019 - 02/2020
- Designed 3D and 2D drawings of circuit protection products and inductors with SolidWorks
- Gained experience with the manufacturing process and usage of electronic parts
Teaching Assitant Experiences
- CMPT 410/726: Machine Learning, Simon Fraser University, 2026 Fall
- CMPT 419: Robotic Autonomy, Simon Fraser University, 2026 Spring
- CMPT 410/726: Machine Learning, Simon Fraser University, 2024 Spring
Research
I. Generalizable and Socially-Aware Human Prediction
Predicting where people will move next is foundational to safe, human-aware autonomous systems, from self-driving vehicles yielding to pedestrians to mobile robots navigating crowded corridors.
Two challenges largely determine whether a prediction method can be deployed in the real world: whether it generalizes beyond the scenes it was trained on, and whether it is socially aware, correctly representing how people influence one another. These two are seldom considered jointly.
In my PhD depth report, "Toward Generalizable and Socially-Aware Human Prediction: A Survey", I review the human trajectory prediction literature along three axes:
1. Generalizability: the leave-one-out (LOO) evaluation policy, online model-update methods that adapt after deployment, and Inverse Reinforcement Learning (IRL), which recovers reward functions that transfer to novel scenes.
2. Social Awareness: physics-based models (Social Force, RVO), game-theoretic formulations, social pooling, attention-based methods (graph attention and transformers), and social-interaction-annotated datasets such as TrajNet++ and JRDB-Social.
3. Evaluation Metrics: displacement-based metrics (ADE/FDE, minADE/minFDE) versus probabilistic and calibration-oriented metrics (NLL, cross-entropy, reliability) that better reflect real-world usefulness but are rarely reported.
Key finding: no existing method uses IRL to recover the rewards behind different types of human social interaction. Combined with the recent interaction-annotated datasets, this is a promising direction toward human prediction that is both generalizable and socially aware.
II. Leaking Corner Issue
1. Admissible Control:
Hamilton-Jacobi reachability analysis is a useful tool for generating reachable sets and corresponding optimal control policies, but its use in high-dimensional systems is hindered by the "curse of dimensionality." Self-contained subsystem decomposition is a proposed solution, but it can produce conservative or incorrect results due to the "leaking corner issue." This issue arises from the inexact decomposition of the target set and inconsistencies across the computed control policies for each coupled subsystem.
In this work, we define and resolve this issue by introducing the notion of an admissible control set that enforces consistent control actions across the coupled subsystems. Our method efficiently computes exact reachable sets and the corresponding optimal control policy for self-contained subsystems with a decomposable goal (or failure) set. We also provide conservative under-approximations for goal (or failure) sets with inexact decomposition. In this conservative case, a local update method in the full dimensional space can be applied to recover exact results.
By calling the issue ``leaking corner issue'', we mean that it is observed that the incorrectness happens at the corner.
In this work, we have the theoretical proof that it only exists in the corner.
Besides, a local updating procedure is proposed to recover the exact reachable set, which guarantees fast computation and correct result.
III. 3D Catheter Reconstruction from Image Sequence
How to find the actual 3D shape of a catheter with a low-resolution endoscopic camera?
In this project, a machine-learning based method is used.
The shape of the catheter is assumed as a Bezier Curve with 3 control points.
With the image sequence, the 3D shape information can be optimized to the true value.
This projected is coded with pytorch, and the code can be found at CatheterControl .
IV. Teleoperation for Catheter Control
The multi-shape advantage of continuum manipulator is crucial in minimum invasive surgery.
In hope for easier usage when implementing surgeries, this project is initiated.
In the following, it shows the mapping relationship between the haptic device and the motors. These motors are aimed for driving the cables which are the director of the manipulator's shape.
This projected is coded with ROS using C++ language
Awards
- Best Design Award, "Harting Cup" Undergraduate Science and Technology Innovation Competition, Tongji University, 2018
