About

I am Yuqiang Yang, a master student at South China University of Technology, supervised by Prof. Chenguang Yang. My research interests are to explore and exploit the potentiality of various robots, such as mobile manipulator, autonomous car, humanoid and multicopter, to perform tasks autonomously and efficiently in unstructured environment.

Specifically, I have explored and realized the possibility of allowing the mobile manipulators to efficiently pick and place in a wholebody manner while avoidance collision smoothly in cluttered dynamic environment full of chairs, tables and shelves. Besides, I am working towards the low-cost and accurate state estimation (Visual Inertial Odometry) , mapping (Occupancy Grid Map and Euclidean Signed Distance Field), planning and control (SE3) framework for quadrotor in complex environments. For autonomous car with non-holonomic constraints, I investigated the real-time perception, mapping and collision-free motion planning to follow and serve the target person well. In the field of robotics learning, I study how to utilize the Nvidia Isaac Lab environment for Unitree H1 humanoid locomotion task. I have sufficient engineering experience on deploying algorithms in various robotic scenarios such as environmental reconstruction, household service and car racing, etc.

Education and Training

Research Experiences

Accepted to IEEE Transaction on Industrial Electronics

Advisor: Dr. Chen Chen, Dr. Zehui Meng

• Developed a novel hierarchical topology-guided searching method to find whole-body kinodynamic trajectories in real time (≈30ms), striking good balance in the success rate and the efficiency.

• Derived and implemented a whole-body trajectory optimization method which naturally integrates the planning and control module (IPC) in an augmented Lagrangian differential dynamic programming (AL-DDP) form.

• Utilized a real-time constructed ESDF map and a novel sphere decomposition for mobile manipulator to achieve one-shot collision detection against dynamic obstacles in batch.

• Conducted extensive experiments to demonstrate the effectiveness of our framework in planning and executing whole-body collision-free trajectories with a high success rate (≈80%) and a excellent tracking accuracy among highly dynamic obstacles.

Accepted to 2023 IEEE Conference on Systems, Man, and Cybernetics

Advisor: Prof. Chenguang Yang

• Proposed a whole-body learning from demonstration framework through Gaussian Process, which endows the mobile manipulator’s skill learning process with features of coordination working and disturbance rejection, after just a few human demonstrations.

• Devised an efficient kinesthetic teaching method based on the weighted least-norm (WLN) inverse kinematics solution and an admittance controller, which facilitates human users to guide the mobile manipulator to perform tasks.

• Implemented a model-based and robust door opening algorithm for mobile manipulator in a coordinated whole-body manner, even under random disturbance.

Project Experiences

Locomotion in complex terrain through reinforcement learning in Isaac lab [Video]

• Designed and fine-tuned the reward function for humanoid locomotion. Then implemented proximal policy optimization algorithm (PPO) to train the humanoid locomotion from curriculum.

• Sim-to-Sim transfer the learned policy into different photo-realistic environments of Nvidia Isaac Sim. Demonstrate robust locomotion ability of the RL policy for Unitree H1.

Risk-aware contingency motion planning under uncertainties for Automated Valet Parking(AVP)

• Realized global topological path searching towards the point of interest(POI). Identified semantic elements on the route (ramp, corner, etc.) and implemented global speed scheduling based on the user preferences and kinematic constraints.

• Planned the lateral path for ego to bypass different agents (static obstalces, upcoming cars, etc.). Generate multiple safe corridors based on vonorio and get the reference polyline. Then we construct second-order continuous SplineGrid from bypass boundary to iteratively optimize a smooth path.

• Modelled the scenario risk considering the uncertainties from perception, prediction and game. Then use iLQR to optimize an interactive longitudinal speed in S-T state space.

• Studied the contingency speed planning under multimodality of prediction. Use tree branch based iLQR to optimize the coupled trajectories in different scenarios.

Low-cost and efficient location, mapping, planning and control for multicopter in embedded system [Video1, Video2, Video3]

Advisor: Dr. Chen Chen; Dr. Zehui Meng; Prof. Fei Gao

• Improved Vins-Fusion by replacing front-end feature extraction with a learning-based feature and enhancing back-end optimization with a QR-based dimensionality reduction algorithm. Boosted the accuracy and robustness greatly in low-performance chip.

• Implemented a low-cost efficient occupancy grid map (OGM) updating algorithm based on incremental inflation and spatial-temporal sliding windows, which enables a extremely fast occupancy information update in large environments.

• Utilized the robot-centric ESDF, whose values are only lazily evaluated in the local frame for extraordinarily small calculations, to formulate the collision cost in the back-end optimization of model predicted contour control (MPCC).

SE3 planning and control for multicopter to cross narrow gap [Video]

Advisor: Prof. Fei Gao

• Constructed the safe flight corridor (SFC) including a series of mutually-consecutive convex polytopes to represent the free space of the quadrotor.

• Utilized MINCO to parameterize the trajectory in spatial-temporal space and jointly optimize a crossing-gap trajectory for better smoothness, using L-BFGS algorithm.

• Identified the quadrotor dynamics parameter for a accurate thrust mapping w.r.t the battery voltage and control duty cycle. Then finely tuned the controller to track the large-attitude trajectory to cross the narrow gap.

Pedestrian following and collision avoidance with spatial-temporal optimization for differential car [Video]

Advisor: Dr. Chen Chen. Dr. Zehui Meng. Prof. Fei Gao

• Proposed a multi-level hybrid A* algorithm to find kinodynamic trajectory to the future position (predicted by EKF) in a certain horizon,jointly considering the object occlusion, uncertainty and collision simultaneously.

• Filtered the object pointcloud of OUSTER lidar based on the bounding box generated from the visual perception and registered the remaining pointcloud of Fast-Lio2 as obstacles.

• Optimized the trajectory parameterized by MINCO spatially and temporally using L-BFGS. Then implemented a low-level MPC controller to accurately track the trajectories under kinematics constraints and communication delay.

Self-balanced racing car with wireless charging capability [Video]

Advisor: Dr. An Chen

• Designed and implemented the adaptive wireless-charging algorithm (PD) to quickly charge (≈30W) the super-capacitors mounted on the car.

• Tuned a cascade position, velocity, attitude and angular controller for a two-wheeled car to race on the track and pass various elements (including circle, slope, crossroads, etc) correctly and steadily.

• Finished the race with the fifth national place in 23.8s and won the first prize.

Wholebody pick-and-place for mobile manipulator [Video]

Advisor: Dr. Chen Chen

• Trained GGCNN for the 6-D perception of the objects, whose inputs are camera pointcloud and outputs are the grasp quality, width and orientation.

• Utilized OSQP to solved the dynamic-weighted QP problem, which jointly considers the manipulability, energy, manipulator orientation and path tracking to achieve coordinated picking-and-placing in wholebody manner.

Honor & Awards

• National Scholarship
  2021-10
• Scholarship of Guangzhou Automobile Group Co., Ltd
  2020-10
• National Scholarship
  2019-10
• The first prize of the National University Students Intelligent Car Race
  2020-08
• Meritorious Winner of Interdisciplinary Contest In Modeling (ICM)
  2021-03

Internship

Skills

Programming:

Python, MATLAB, C/C++, PyTorch, Pybullet, Airsim, Embedded System

Robotics:

Wholebody control, Peception and mapping, Convex Optimization, Admittance/Impedance Control, Gravity Compensation, teleoperation

Hardware Experience:

Multicopter, Diablo, Franka, UR, Mobile Manipulator, Robotiq 2F85, Vicon, Touch X, ATI sensors, STM32