top of page

ProJect

Minimally-Intrusive
Navigation
in Densely-Populated Pedestrian Flow

Dedicated to researching robotic navigation tasks in dense crowds. Proposed the concept of Minimal Intrusive Navigation, which is composed of two levels of disturbances in pedestrian flow. Aimed to address the "freezing robot problem" in crowd navigation. Qunitify the algorithm on both macro and micro level.

Task  Agnostic and  EgoCentric Reinforcemen  Learning
in Autonomous Driving

Enabled autonomous vehicles to explore effectively, so that they can perform temporally extended actions that reflect specific driving intentions.

Expert-Guided
Motion-Encoding Tree
Search in Autonomous Driving

Incorporated motion primitive methods into Monte Carlo Tree Search (MCTS), which allows searching complexity for long-duration MCTS tasks by decreasing the depth of the tree. Expert policy involvement search enables effective exploration.

图片1.png

LightZero

Participated in the development of LightZero, an open-source algorithm tookit that combines Monte Carlo Tree Search and Deep Reinforcement Learning

DI-drive

Focused on developing decision intellegence platform for autonomous driving simulator. Implemented macro-level functionalities within Metadrive simulator's highway environment; and Implement the MPC into CARLA simulator; Replicated the D-REX algorithm in Drive environment.

Online State-Time
Trajectory Planning 
in Highly  Dynamic CrowdEnvironments

Propose a gradient-based planner over the state-time space for online trajectory generation in highly dynamic environments

Human
Trajectory Prediction

Collected naturally crowded pedestrian datasets from metro stations and plazas in Hong Kong by calibrated lidar and camera system. Used an attention-based social pooling layer to model interactions among pedestrians.

Automatic Recycling System
Airport Baggage Cart

Aimed to address the use of robotic technique to replace humans in performing labor-intensive tasks. Proposed a method and equipment for collecting baggage carts, which enabled automatic collection of the carts. Involved the design of perception, localization, decision making and mechanical modules.

 An Autonomous
Eye-in-Hand robotic
System for Elevator Button Operation

Aimed to address the challenge of autonomous transportation for robots across different scenarios. Propose an autonomous robot system with Eye-in-Hand structure to solve the button operation problem. Includes perception, control and planning tasks.

bottom of page