Projects
Locally Enforced Optimal Control (LEOC)
There have been attempts in reinforcement learning to exploit a priori knowledge about the structure of the system. We propose a hybrid RL controller which dynamically interpolates a model-based linear controller and an arbitrary differentiable policy. The linear controller is designed based on local linearised model knowledge, and stabilizes the system about a target state. Learning has been done on both model-based (PILCO) and model-free (DDPG) frameworks. The overall hybrid controller is proven to maintain the desirable properties of both the linear and arbitrary non-linear policies.
View Paper PMLR Volume 144 View Project Code
MQA: Answering Questions via Robotic Manipulation
We propose a novel task of Manipulation Question Answering (MQA), where the robot is required to find the answer to the question by actively interacting with the environment via manipulation. Considering the tabletop scenario, a heatmap of the scene is generated to facilitate the robot to have a semantic understanding of the scene and an imitation learning approach with semantic understanding metric is used to generate manipulation actions. Extensive experiments have been conducted to validate the effectiveness of the proposed framework.
View Paper on arxiv View Project Code
Raft Protocol Implementation
A distributed key/value service based on the Raft protocol is implemented in Golang as part of MIT 6.824/THU Distributed Systems course work.
View Project Code
Property Guru scraper
This is a tool to subvert real estate agents' incessant refreshing of their old property listings to make them look new. The scraper filters out old listings from relevant searchers, and extracts the genuine new listings.
View Project Code