Home | Zilin Chen's Homepage

7-DOF Robotic Arm Motion Planning Algorithm Development for Humanoid Robots

Tue, 15 Jul 2025 00:00:00 +0000

Project Overview

Core Objective: Develop motion planning algorithms for Xiaomi’s 7-DOF humanoid robotic arm, enhancing motion precision (<0.3mm error), stability, and robustness against payload variations.
Technical Framework:

Theoretical Basis: DH parameters, inverse kinematics (IK), trajectory interpolation
Toolchain: MuJoCo simulation, C/Python co-development
Validation Protocol: Dual verification system complying with ISO 9283 standards (positioning accuracy, path repeatability)

Key Technical Achievements

Kinematics Algorithm Development
- Forward Kinematics: Established DH-based coordinate transformation model with 0.1mm calculation precision
- IK Solver Optimization:
  - Developed universal inverse kinematics interface auto-adapting to URDF configurations
  - Hybrid numerical-geometric solution reducing end-effector positioning error to 0.3mm
- Trajectory Smoothing: Implemented Kalman filtering with B-spline interpolation, decreasing joint velocity fluctuations by 30%
Simulation-Physical Validation
- MuJoCo Environment: Calibrated dynamic parameters (gravity compensation, joint friction models) matching h1_2 hardware specs
- Hardware-in-the-Loop Validation: <1.5% deviation in 50 test trajectories between simulation and physical tests

trajetory planner with filter

enduable test results demo

System Optimization

Three-Core Enhancement Strategy:

Fail-Safe Mechanism: Collision detection triggering 10ms emergency stop
Computational Efficiency: optimized code achieving 1kHz control frequency

Strategic Recommendations:

Engineering: Implement fault injection testing for failure mode analysis
Collaboration: Establish knowledge-sharing platform for URDF optimization

Ultra-Lightweight DEA Drone Flight Controller Development

Tue, 04 Mar 2025 00:00:00 +0000

Since March 2025, I have been collaborating with Professor Huichan Zhao to develop an innovative flight control system for an ultra-lightweight drone utilizing soft artificial muscles. The drone weighs only 23 grams and employs Dielectric Elastomer Actuators (DEA) as its primary propulsion mechanism.

Project Overview

This research focuses on overcoming the unique challenges of controlling drones powered by soft artificial muscles. Unlike traditional motor-based systems, DEA actuators offer silent operation, high energy efficiency, and biomimetic movement patterns, but require specialized control approaches due to their nonlinear electromechanical properties.

Flight Controller Development

I am designing a custom flight controller architecture specifically optimized for DEA-driven drones:

Implemented real-time control algorithms accounting for DEA hysteresis and viscoelastic behavior
Developed adaptive PID controllers with gain scheduling for voltage-to-strain conversion
Integrated sensor fusion for IMU data processing with Kalman filtering

Simulator Implementation

To enable safe testing and rapid prototyping, I built a physics-based simulator featuring:

High-fidelity DEA actuator models with electromechanical coupling dynamics
Aerodynamic modeling for ultra-lightweight frames
Real-time visualization of drone state and actuator deformation

Technical Challenges

The project addresses several key research challenges:

Nonlinear Control: Compensating for DEA’s voltage-dependent strain response and creep effects
Weight Constraints: Implementing full control stack within 23g total system mass

Current Progress

As of today, we have achieved:

Successful closed-loop attitude control in simulation with <5° steady-state error
Preliminary flight tests in mujoco simulation, demonstrating basic stabilization capabilities

[Project repository link to be added upon publication]

Skills

Sat, 01 Mar 2025 00:00:00 +0000

Academic skills

Programming language

I am familiar with a number of programming languages:

C (including C99/clang/gcc, etc.)
C++
CMake
Python
MATLAB

I’m also a user for other languages:

Java

Software

I know how to use several useful tools:

AutoCAD
SolidWorks
Multisim
Ansys

Musical Instrument

I am a piano player, also a double bass player. I know some other instruments as well, such as cello.

I earn myself some title, such as:

Central Conservatory of Music Piano Amateur Player, Grade 9 (highest grade)
China Conservatory of Music Double Bass Player, Grade 10 (highest grade)
Practical Grade 8 for Piano, ABRSM (highest grade in practical)

I’m also the former Vice President of Tsinghua University Symphonic Orchestra, and the conductor for its second team.

UTAT-ADR Autonomous Drone Racing Research

Fri, 20 Sep 2024 00:00:00 +0000

From September till December, 2024, I joint a team in UTIAS, University of Toronto, which mainly focus on Autonomous Drone Racing.

Under professor Hugh H. T. Liu’s guidance, I mainly focus on building a simulator for the drone, also planned the yaw angle.

Research Overview

Human pilots can use their own talent to controll drones to fly rapidly through gates, and the goal of the whole team is to use computer-based autonomous pilot to guide the drone.

My Contribution

Building a simulation system

The team previously used Betaflight to be its flight controller. However, to make full use of AI, we need to simulate how the drone behaves. So, developing a simulator according to Betaflight’s original code is very important.

I read all the lines from Betaflight, and extracted the controller code and test it. Also, I tested the input signal and output RPM curve.

This task is not a hard one, but it took me a lot of time to get to know how Betaflight work. As an open-source code, it’s super long and hard to read.

The following is the code structure of Betaflight that I organized：

This one is a demo of the part I contributed for the simulator:

Yaw planning

The racing drone we built is using computer vision to guide and locate itself. The drone need to look at the “gate” to get to know where it is.

My work is to help the drone get better vision to have a more accurate position.

The following is a demo of the yaw-planning. The red arrow is the yaw angle of a world-champion level human-pilot, while the blue arrow is the planned one. The black line is the gate.

Click here to see the orinigal code.

Intelligent Autonomous Guided Car

Mon, 01 Jul 2024 00:00:00 +0000

During one of our summer semester course, I led a team of three in designing an autonomous car-shaped vehicle capable of transporting objects from start to finish. The vehicle have integrated advanced features including line-following, obstacle navigation, destination location, and Bluetooth connectivity for remote control.

I acquired proficiency in utilizing PID control algorithms to fine-tune input parameters for precise motor control. I utilized an STM32 microcontroller for vehicle control and Open MV for image capture (computer vision), trajectory planning, and dynamic output management.

Task

Our vehicle is asked to finish two task automatically:

line-following: pick up the stuff and carry it, move along the line and then place the stuff down at the end point
path-finding: knowing the start-point and the end-point absolute position in the real world, but have a lot of obstacles in the path. The vehicle is asked to plan a suitable way for itself to find the end and put the stuff down.

Hardware

We use STM32 as the main controller, controlling all the motors, servos, and send and grab data from bluetooth, openMV and Gyro (We use JY901S).

In the main function, read OpenMV, Bluetooth, and IMU data in a loop with 5ms as one unit. One loop takes 20ms (the remaining 5ms is used to process all the above data). Refresh motor and servo control data in each loop.

Detail issues:

Interrupt conflict: prioritize encoder interrupt to ensure PID stability.
Abnormal data: retain the last received data without refreshing.
Check bit: avoid incorrect vehicle state control caused by accidental error transmission of data, enhance fault tolerance.

Line following

The vehicle can follow the line automatically. Here is an video when we are still testing:

您的浏览器不支持视频标签。

Click here if not working.

Path Finding

Here is one video while testing:

您的浏览器不支持视频标签。

Click here if not working.

Here is another one:

您的浏览器不支持视频标签。

Click here if not working.

Research on Dual-arm Collaborative Robot Working Space

Tue, 12 Mar 2024 00:00:00 +0000

From Febuary till December, 2024, I joint a team of four to work on a dual-arm collaborative robot system.

Under assistant professor Ze Wang’s guidance, I mainly focus on finding the work space of the robot.

Research Overview

Advisor: Ze Wang, Professor and Assistant Dean at School of Mechanical Engineering, Tsinghua University

The main focus of this study is to use dual-arm collaborative robot system to enhance the stiffness while working. Most dual-arm system do not have a rigid connection between the two arms. We are developing a system which have a rigid connection, which is the main innovation.

We designed three different configuration for the dual-arm system, as shown below:

Motion Trajectory Planning Research
- Summarized current research progress on motion trajectory planning of two-arm collaborative robots.
- Identified key areas for improvement and proposed innovative solutions.
End Effector Stability Enhancement
- Improved the stability on the end effector of a two-arm collaborative robot.
- Implemented techniques to ensure precise and consistent performance.
Workspace and Collision Analysis
- Analyzed the workspace and constraints of the robot.
- Developed strategies to prevent collisions between the two arms of the collaborative robot.
Robot Configuration Design Collaboration
- Collaborated with a team of four to design different robot configurations.
- Focused on improving stiffness and overall structural integrity of the robots.

My Contribution

I mainly focus on the working space of this system. Unlike most dual-arm robot system, ours need to consider the problem of two robotic arms interfering with each other.

Using MATLAB to do simulating work, I developed a programme using Monte Carlo method to find the available workspace.

One example is shown below:

The red dots is the available dots, where the robot can reach under the strict rigid connection.

We can see the slice on the plate of z = 0.2:

Below is what is looked like on this plate:

Click here to find the original code