Future is Automation
Transfusor: Transformer Diffusor for Controllable Human-like Vehicle Trajectory Generation
Goal: Generate human-like lane-changing trajectories following certain condition
Paper coming soon
1. Develop a generative model with only a limited number of training samples
2. Achieve robust and realistic lane-changing trajectories with different aggressiveness level
3. Greatly enrich the virtual simulation test scenarios for AV testing
LiDAR-Based Cooperative Relative Localization
Conference: IV Symposium (June 2023, coming soon)
Goal: Reduce relative localization with LiDAR points registration
1. Develop a robust localization framework for multi-CAV cooperative perception
2. Achieve fast and efficient relative localization and reduce 80 % GNSS errors
3. Save communication bandwidth by only transforming a small portion of points
Online Media 23.movModel-based RL framework for collision avoidance
Poster: TRB 2023
1. Optimal control + Reinforcement learning to avoid the collision when others are "at fault"
2. Trajectory predicting model (DL based)
3. Dynamic trajectory planning and optimization
Spatio-weighted information fusion and DRL-based control for connected autonomous vehicles
Conference version: 23 rd IEEE ITSC (September 2020). Poster: TRB 2021
Journal: Transportation Research Part C: Emerging Technology May 2021
Goal: faster, safer, comfort lane-changing decisions
1. Data fusion framework for both local information (from sensors) and global information (from connectivity devices) ----> dynamic shape input
2. Q learning as a lane-changing decision processor
3. Optimal connectivity range analysis.
Rayne - OBS-animation2 流畅版.mp4A DRL-based Multiagent Cooperative Control Framework for CAV Networks: a Graphic Convolution Q Network
Journal: Computer‐aided Civil and Infrastructure Engineering (April 2021):
Poster: TRB 2021; Full paper: Arxiv
Goal: Cooperative lane changes to merge out the ramps
1. GCN-based information fusion block to enforce communication and cooperation between vehicles
2. Combine GCN and DQN into an end-to-end multi-agent decision processor to control the CAVs lane changing decisions for a road segment.
Image transformer for explainable autonomous driving system
Conference version: 24th IEEE ITSC (September 2021)
Poster: TRB 2022
Journal: Journal of Intelligent and Connected Vehicles
Image transformer-based end-to-end computer vision model.
Generating driving decisions and explanations.
Achieve SOTA with significantly superior performance and lower computational cost compared to the benchmark model.
CAV and Multi-Agent Reinforcement Learning for Mitigating Highway Bottleneck Congestion
Abridged version: Arxiv
Journal version: Transportation Metrica Part A.
Goal: Achieve variable speed limit to reduce bottleneck congestion using CAVs
1. Reinforcement learning + Graphic neural network
2. Bottleneck congestion mitigation with variable speed limit (VSL)
Convex optimization based path planning for autonomous vehicles
1. Implement the Convex Feasible Set algorithm for real-time path planning
2. Compare Convex Feasible Set with SQP, iterative SQP for performance evaluation