Portfolio

This page shows a portfolio of projects that have been done or doing since my graduate study at The Ohio State University. They mainly consist of pedestrian-related problems or issues in the autonomous driving. The topics ranges from perception, prediction, motion planning and control of autonomous driving tasks that require interaction with pedestrians.

Below is just a list of projects, detailed description is coming soon.

Online Pedestrian/Object Detection

• Image-based 2D object/pedestrian detection by deep neural network (ResNet-34).
• Pointcloud-based 3D object detection by rule-based segementation.
• Calibration of a combined camera and LiDAR sensor system.

Interactive Pedestrian Motion Modeling for Traffic Scenario Simulation

• Social force based modeling of pedestrian/crowd motion considering vehicle effect.

Pedestrian Trajectory Prediction

• Neural network (LSTM) based interactive pedestrian trajectory prediction.
• Reachable set analyis of pedestrian behavior.

Pedestrian Dataset

• Created two pedestrian trajectory datasets that covers multi-pedestrian interaction with vehicles:
  • CITR Dataset: https://github.com/dongfang-steven-yang/vci-dataset-citr
  • DUT Dataset: https://github.com/dongfang-steven-yang/vci-dataset-dut
• Standardized different pedestrian trajectory dataset for better benchmark configuration.

Vehicle Motion Planning in Mixed Environment

• Local trajectory planning (lattice planning) for obstacle bypassing based on optimizing high-order polynomials in the Frenet frame.
• Path planning in the unstructured environment by A*, modified RRT, and spline fitting.

Longitudinal Speed Control of Vehicle Interacting Crossing Pedestrians

• Longitudinal speed regulation by model predictive control (MPC) with the prediction of moving obstacles by reachability set analysis.